How SM-2 Scheduling Works (and Why Your Flashcards Come Back Right Before You Forget)

12 Jun 2026

Every serious flashcard app — Anki, SuperMemo, Word Holder — is built on the same core insight: memory has a schedule, and you can ride it. This post explains the algorithm most of them descend from, SM-2, in plain language: where it came from, how it decides when you'll see a card again, and why that timing feels uncanny once you've used it for a few weeks.

The forgetting curve

In the 1880s, Hermann Ebbinghaus memorized lists of nonsense syllables and tested himself at intervals. The result is the famous forgetting curve: memory decays fast at first, then ever more slowly. You lose most of a new fact within days — but each time you successfully recall it, the curve flattens. The second forgetting is slower than the first, the third slower still.

That suggests an efficient strategy: review a fact just before you'd forget it, each time at a longer interval. Review earlier and you waste time on things you still know; review later and you're relearning from scratch. The hard part is knowing where "just before you forget" falls for each individual fact — which is exactly what SM-2 estimates.

SM-2 in plain language

SM-2 was developed by Piotr Woźniak for the first computer versions of SuperMemo in the late 1980s (the "SM" stands for SuperMemo). For each card it tracks two things:

An interval — how long until the card is shown again.
An ease factor — a per-card difficulty estimate, starting at 2.5.

The schedule works like this:

First successful review: see the card again in 1 day.
Second: 6 days.
After that: each new interval is the previous interval × the ease factor. With the default 2.5, that's roughly 15 days, then 37, then 94, then 235 — under a handful of reviews and the card is coming back at multi-month intervals.
Your answer adjusts the ease factor. Recall it easily and the ease factor grows — intervals stretch faster. Struggle and it shrinks — that card comes back on a tighter leash. (It never drops below 1.3, so no card gets punished forever.)
Fail a card and its interval resets: you'll see it again tomorrow and rebuild from short steps. The card "lapses" but the algorithm remembers it was once known — relearning is faster than learning.

That's the whole machine. A few multiplications per card, no neural networks — and it captures something real: each successful recall earns an exponentially longer holiday before the next one.

Why cards come back "right before you forget"

The eerie experience every SRS user reports — I was just about to forget that one — isn't magic. The ease factor is a feedback loop: every answer you give nudges the card's schedule toward the edge of your memory. Cards you find easy drift out to long intervals until they, too, become borderline. Cards you find hard get pulled in until they stop being hard. The steady state, by construction, is that most reviews land near the forgetting point — which is also the point of maximum learning efficiency.

That's also why a daily ten-minute session can maintain thousands of words: at any moment, almost all your cards are far away on long intervals, and only the few that are due — the ones near their forgetting point — appear today.

After SM-2: SM-20 and FSRS

SM-2 dates from the late 1980s, and research didn't stop there. SuperMemo's recent algorithms model memory stability and retrievability separately, and its 2026 release ships SM-20. Anki's modern optional scheduler, FSRS — which you turn on in deck options — fits a memory model to your personal review history and measurably beats stock SM-2 at predicting recall. If you enjoy that frontier, our comparison of spaced-repetition tools covers which tools run what.

But the refinements are smaller than they sound for everyday vocabulary learning. The dominant factors are still the SM-2 fundamentals: expanding intervals, per-card difficulty, review-on-time. A learner with a mediocre algorithm and a sustainable habit beats a learner with an optimal algorithm and a dead deck — which is why the workflow around the algorithm (who makes the cards, where reviewing happens) matters more than the scheduler version.

Where you'll meet SM-2 today

Anki's classic scheduler is an SM-2 derivative with decades of community tuning (its modern opt-in scheduler, FSRS, builds on the same lineage). Countless smaller apps implement SM-2 straight from Woźniak's original description.

Word Holder uses SM-2 too, with one design decision worth explaining: the algorithm is completely hidden. Cards create themselves from the words you look up while reading, and at review time you see a fill-in-the-blank sentence and four buttons — Again, Hard, Good, Easy. Those four answers are the standard SM-2 quality grades; the ease factors and intervals update behind the scenes, with no settings to tune. You get the classic algorithm's full effect — words returning right before you'd forget them — without ever seeing a number.

If you've never felt an SRS work on words you actually chose, read a short story and tap a few words — no account needed. The schedule starts immediately; come back tomorrow and SM-2 will have your first reviews waiting.

Corrections welcome — spotted something wrong or out of date? Let us know.