FSRS, the scheduling algorithm behind modern Anki

R3CK0N's Flashcards section schedules reviews with FSRS — the Free Spaced Repetition Scheduler, developed by the open-source open-spaced-repetition research project and now the default algorithm in Anki itself. It replaced the older SM-2 method because it models memory more directly, and it can explain itself in plain language even though the math underneath is not simple.

Three numbers per card

Every card FSRS has ever scheduled carries three running estimates, updated after each answer:

Stability

Stability is roughly "how many days until you'd have a 50/50 chance of remembering this." A brand-new fact might have a stability of half a day; a fact you've answered correctly a dozen times might have a stability of years. Every successful review pushes stability up; every miss pulls it back down. This one number is what actually drives how far out the next review gets scheduled.

Difficulty

Difficulty is a 1-to-10 rating of how hard a card is for you, independent of how long you've known it. It grows when you press Again or Hard, and eases back down over repeated easy passes. Two cards can have identical stability and still get different intervals, because the harder one is treated more cautiously going forward.

Retrievability

Retrievability is the live output: given a card's current stability and how long it's been since you last saw it, what's the actual probability you'd get it right right now? It starts near 100% right after a review and decays continuously — this is the "forgetting curve" for that specific card, and it's what the scheduler is really trying to manage.

Desired retention: the dial you control

FSRS doesn't pick intervals in a vacuum — it targets a retrievability you choose, called desired retention, defaulting to 90% and adjustable per deck between 70% and 97%. Set it higher and the scheduler shows you cards more often, trading review time for a lower forgetting rate. Set it lower and reviews get sparser but you'll fail more of them. There's no universally correct value; it's a trade-off between total time spent and how often you're willing to be wrong.

Why two cards you learned on the same day get different schedules

This is the part that looks strange coming from a fixed-interval system: two cards added on the same day, both answered "Good" every time, can end up months apart in their schedule. That's the whole point. One might have turned out easy — each pass, its stability climbed fast, so its intervals stretched quickly. The other might be a harder fact that took a couple of early misses to settle down, so its difficulty sits higher and its intervals grow more cautiously. FSRS is tracking each card's own memory curve, not applying one formula to everything.

The 21 parameters, in one sentence

FSRS-6 has 21 numeric parameters that control exactly how stability and difficulty move after each grade — how big a jump Easy gives you, how much a lapse costs, how quickly difficulty eases back down, and more. In one sentence: those 21 numbers can be fit to your own review history so the model reflects how your memory actually behaves, rather than assuming everyone forgets at the same rate. R3CK0N starts every deck with the same sensible defaults; a deck can hold its own tuned set if you bring one in from an Anki import that already had them optimized.

Learning steps: the short-term ramp before FSRS takes over

A brand-new card, or one you just failed, doesn't go straight into the stability/difficulty model — it first runs through a short sequence of learning steps, minutes apart (a minute, then ten, by default), so a fact you just saw gets reinforced quickly while it's fragile. Only once it graduates past the last step does it enter the FSRS-scheduled review pool with a real interval in days. This mirrors how Anki itself splits the problem: fast repetition for material you haven't retained at all yet, and the slower, longer-range model for material you're maintaining.

What this means day to day

You never see stability, difficulty or retrievability numbers while you're studying — you just see a card and press Again, Hard, Good or Easy. FSRS is the machinery translating that grade into "come back in 40 minutes" or "come back in four months." The Stats panel does surface the underlying numbers, though, as histograms of stability, difficulty and retrievability across your whole collection, if you want to see how the model rates what you know.

Frequently asked questions

Is FSRS the same as Anki's SM-2?

No. SM-2 used a single "ease factor" per card and simpler fixed multipliers. FSRS models stability, difficulty and retrievability separately and fits them from real review data, which is why modern Anki switched to it as the default.

Do I need to understand any of this to use Flashcards?

No — the only settings you'll normally touch are desired retention and how many new cards to add per day. Everything else runs in the background.

What does "FSRS-6" mean — is there a difference from earlier versions?

FSRS has gone through several revisions as the open-spaced-repetition project has refined the model; 6 is the current generation, with 21 parameters instead of earlier versions' fewer or differently arranged ones.

Does importing from Anki bring my FSRS data with it?

Yes, when you import a Package (.apkg) with scheduling kept — see the Anki import guide. Each card's stability and difficulty come across so reviews continue where Anki left off.