Spaced repetition works because memory fades on a predictable curve and every successful recall flattens that curve. A self-scheduling flashcard system records how well you remembered each card, then sets the next review just before you would probably forget it, so the gap grows after each success and collapses after each failure. Reviews spread out this way produce more durable memory than the same reviews packed into one session, which is why distributed practice earns a high utility rating in the research.
What does the forgetting curve actually show?
Hermann Ebbinghaus measured his own forgetting in the 1880s, and in 2015 Jaap Murre and Joeri Dros replicated the experiment with one subject learning nonsense syllables for 70 hours. Their savings scores, the share of the original learning effort saved when relearning, were 0.585 after 20 minutes, 0.422 after one hour, 0.220 after one day, 0.122 after six days and 0.041 after 31 days.
Two things matter for a student. Most of the loss happens in the first day, so a review on day one is worth far more than a review in week three. And the curve is not smooth: the authors report a probable jump upward starting at the 24 hour point, which they connect to research on sleep consolidating memory.
A self-scheduling system reads this curve as a budget. It spends a review where the curve is about to drop below your target, and it skips reviews where recall is still near certain.
Why does spacing beat cramming?
The largest synthesis is the 2006 review by Cepeda, Pashler, Vul, Wixted and Rohrer, which covered 839 assessments of distributed practice across 317 experiments in 184 articles. Its central finding is that the gap between sessions that produces maximal retention increased as the retention interval increased. In plain terms, the further away the test, the wider your gaps should be.
The follow-up 2008 study by the same group put numbers on it with more than 1,350 participants who learned facts, reviewed after gaps of up to 3.5 months, and were tested up to a year later. At any test delay, lengthening the gap first improved and then reduced final performance. The optimal gap was roughly 20 to 40% of the test delay for a test one week out, and roughly 5 to 10% for a test one year out.
Cramming sits at the wrong end of both findings. A gap of zero is the worst point on the curve for any test that is more than a few minutes away.
Why flashcards rather than rereading?
Spacing decides when you study. Flashcards decide how, and the how is retrieval: pulling the answer out of memory instead of looking at it again. In a 2006 experiment by Roediger and Karpicke, students who restudied a passage did better than students who were tested on it when the final test came five minutes later. On tests two days and one week later, prior testing produced substantially greater retention than restudying.
The uncomfortable part of that paper is the confidence data. Repeated studying increased the students' confidence in their ability to remember the material, while it produced the weaker delayed result. Rereading feels productive precisely because the text is in front of you.
The 2013 review by Dunlosky and colleagues rated ten study techniques. Practice testing and distributed practice were the only two rated high utility. Rereading, highlighting, summarization, the keyword mnemonic and imagery for text were rated low utility. A spaced flashcard deck combines the two high utility techniques in one habit.
How does a flashcard app decide when to show a card again?
Every scheduler answers the same question: given how well you just answered, how many days until the next review? The rules differ in how much they adapt to you.
| Scheduler | Origin | How the gap is set | Reads about you |
|---|---|---|---|
| Leitner boxes | Sebastian Leitner, 1972 | Correct cards move to a box reviewed less often; wrong cards return to box one | Only pass or fail |
| SM-2 | Piotr Wozniak, 1990 | Fixed first gaps, then each gap is the previous gap times an ease factor that moves with your grades | A 0 to 5 grade per card |
| FSRS | Open source, used by Anki | Models difficulty, stability and probability of recall per card, then schedules to a target recall rate | Your full review history |
The Leitner system was published in 1972 in a book on how to learn to learn. Cards are grouped by how well you know them, correct answers promote a card to the next group with a longer review interval, and a wrong answer sends it back to the first group.
SM-2, published by Piotr Wozniak in 1990, is the algorithm most flashcard apps descend from. The first interval is one day, the second is six days, and every interval after that is the previous interval multiplied by an ease factor. Each card starts with an ease factor of 2.5 that can never fall below 1.3, and you grade every answer from 0 (complete blackout) to 5 (perfect response). Any grade below 3 restarts the card from the beginning. The Anki manual gives the shape of the resulting schedule: a card seen today might return in 3 days, then 15 days, then 45 days.
FSRS replaces the single ease factor with three quantities per card: difficulty on a 1 to 10 scale, stability, defined as the interval at which the probability of recall drops to 90%, and retrievability, the probability of recall right now. Earlier versions modeled the forgetting curve as R = 0.9^(t/S), so recall is exactly 90% when the elapsed time equals the stability. The scheduler then picks the next date at which retrievability will hit your target. In Anki's deck options that target is called desired retention, defaults to 90%, and the manual warns that above 90% the workload increases very quickly and above 97% it can be overwhelming. Lower it and you do significantly fewer reviews per day but forget more cards.
How should you set it up for an exam?
Start the deck the day you get the material, not the week before the exam. The 2008 Cepeda study puts the best first gap at about a fifth to two fifths of the time until a test one week away, and a far smaller share for a test a year away, so a deck that is a month old by exam day has already banked its most valuable reviews.
Write cards that force recall of one fact each, and grade yourself honestly. SM-2 treats a grade of 2 as a failure, FSRS lowers stability on a lapse, and both are only as accurate as the grades you give them. A deck of "I sort of knew that" marked as correct will schedule reviews you cannot afford to miss.
Let the tool own the calendar. If you are building the deck from lecture slides and PDFs, StudyOS turns those files into flashcards with spaced repetition, mock exams, and a readiness score for each exam, with every answer citing the page it came from.
What spaced repetition cannot do
A scheduler protects memories that already exist. It does not create understanding, and the 2013 review rated techniques that build understanding, such as self-explanation and elaborative interrogation, as moderate utility because they lack classroom scale evidence rather than because they fail. Use cards for the facts, definitions, formulas and steps you must reproduce, and use problems and explanation for the rest.
It also cannot fix a bad card. Ambiguous questions, cards with three facts on the back, and cards copied from a summary you never understood all get scheduled with the same confidence as good ones.
Frequently asked questions
How long does it take for spaced repetition to work?
The first review pays off within a day, because the 2015 replication of Ebbinghaus shows that most forgetting happens in the first 24 hours.
The larger benefit over cramming shows up on delayed tests: in the Roediger and Karpicke experiment the advantage of retrieval was visible at two days and one week, not at five minutes.
Is spaced repetition better than rereading my notes?
On any test more than a few minutes away, yes.
Rereading was rated low utility in the 2013 Dunlosky review, and retrieval practice produced substantially greater retention than restudying on delayed tests, even though restudying raised students' confidence more.
Spaced flashcards combine retrieval with the high utility technique of distributed practice.
What is the difference between SM-2 and FSRS?
SM-2 uses fixed first intervals of one and six days and then multiplies by an ease factor that moves with your 0 to 5 grades.
FSRS estimates difficulty, stability and probability of recall for each card from your whole review history and schedules to a target retention, 90% by default in Anki, which you can lower to cut the daily workload.
How many cards a day is realistic?
That depends on your desired retention and the size of the deck, not on a fixed number.
Anki's documentation notes that workload rises very quickly above 90% desired retention and can be overwhelming above 97%, so if reviews pile up, lower the target slightly rather than skipping days and letting the schedule drift.