Spaced Repetition: Never Forget What You Learn [2026]

You study for two hours straight. Your friend studies for 20 minutes, four times over two weeks. A month later, your friend remembers twice as much.

Spaced repetition is the technique behind that gap: a system that times each review to land right before you would forget. Hermann Ebbinghaus measured the decay in 1885 and found that roughly two-thirds of freshly memorized material fades within 24 hours, a curve that has held up in modern replications. LearnClash schedules missed quiz questions on this principle, which is why a question you fumble in a duel quietly comes back a week later.

Below: the forgetting curve, the review intervals the research actually supports, why cramming feels productive while it quietly fails, and the three-stage system I ended up building instead of an infinite one. Test your memory on any topic →

What Is Spaced Repetition?

The oldest working version of the idea fits in a shoebox. In 1972, the German science journalist Sebastian Leitner published So Lernt Man Lernen (“How to Learn to Learn”), which taught a card-box method: answer a flashcard correctly and it moves to a box you review less often, miss it and it drops back to the first box, the one you work through most often. No software, no algorithm. Cardboard partitions did the scheduling.

That is the whole mechanism, then and now. A spaced repetition system tracks how well you know each item and stretches the gap before the next review as your answers improve, so weak items cycle fast while strong items rest. Everything since 1972 has been refinement of that loop; the Leitner system breakdown covers the original five boxes and what became of them.

Leitner ran this on cardboard in 1972. The loop has not changed since; only the scheduling got smarter.

Why Do We Forget? The Forgetting Curve Explained

In 1885, Hermann Ebbinghaus sat alone in his apartment memorizing lists of nonsense syllables, then measured how quickly they leaked away. His savings scores, the share of relearning effort spared: 58% after 20 minutes, 44% after an hour, 33% after a day, 25% after a week, 21% after a month.

Ebbinghaus’s savings scores without review. Each spaced review resets the curve at a higher floor.

One caveat matters. Those figures come from nonsense syllables, close to a worst case for memory, and meaningful material decays along a flatter slope. The shape is what survives scrutiny: in 2015, Murre and Dros replicated the 1885 experiment in a different language, 130 years later, and got nearly identical curves.

The curve has a flip side that traps diligent students specifically. Right after a session, everything feels known. Robert Bjork calls that feeling the illusion of competence, and rereading feeds it, because the brain confuses “that looks familiar” with “I could produce that answer.”

	Recognition	Recall
What your brain does	”That looks familiar"	"What’s the answer?”
Feels like	Easy, confident	Hard, uncertain
Actual retention	Fades in days	Lasts months to years
Triggered by	Rereading notes	Quiz questions, duels

Spaced repetition forces the recall column. It never asks whether something looks familiar; it asks for the answer, after a gap long enough that retrieving it takes real work.

“Only 12 of 271 comparisons of massed and spaced performance showed no effect or a negative effect from spacing, making the spacing effect quite robust.” (Cepeda et al., Psychological Bulletin, 2006)

How Does Spaced Repetition Actually Work?

The story of how a 19th-century notebook observation became working software starts with a frustrated student in Poland. In 1985, a 23-year-old student in Poznań named Piotr Wozniak, fresh out of a molecular biology degree and starting computer science, was drowning in English vocabulary. He ran experiments on himself with paper lists, logging every review and every failure, and noticed that after each successful review the safe gap before forgetting roughly doubled.

His family pooled money to buy him a PC. On December 13, 1987, Wozniak loaded SuperMemo 1.0 onto his DOS machine, and the refined algorithm, SM-2, appeared in his 1990 master’s thesis at Poznań University of Technology. Anki ran on SM-2 for years and now defaults to FSRS, and SM-2 descendants still drive a generation of learning apps.

Modern neuroscience supplies the mechanism Wozniak could only infer. A 2025 study in Communications Biology reported that spaced learning promotes neural integration and replay in the cortex rather than the hippocampus. The distinction matters: the hippocampus is a short-term buffer, and the cortex is long-term storage.

Here is what happens in your brain during spaced versus massed learning:

1. Cramming: material parks in the hippocampus and fades within days.
2. Spaced review: each retrieval pushes the memory into cortical networks that persist for months.
3. Sleep between sessions: slow-wave sleep replays the spaced material and consolidates it. A single cram session skips this step entirely.

There is a catch, and it is the reason so few people stick with the method: spacing feels worse than cramming while you are doing it. Bjork named this desirable difficulty in 1994, and he and Elizabeth Bjork expanded the framework in their 2011 chapter. Students consistently rate massed practice as more effective, then the spaced group outperforms them decisively a week later. The struggle during recall is not interference with the learning; it is the learning.

What Are the Optimal Spacing Intervals?

The optimal gap between reviews scales with how long you need to remember something: Cepeda et al. (2008) mapped it across 1,354 participants and found the gap runs about 20 to 40 percent of a one-week retention horizon and falls to about 5 to 10 percent of a one-year horizon. The shorter the horizon, the larger the share. For a test next week, review after 1 to 2 days. For knowledge you want in a year, the first review belongs 3 to 4 weeks out.

The study ran gaps up to 3.5 months with final tests up to a year later, and the optimal gap scaled with the retention goal in every condition. No fixed calendar can serve a Friday quiz and a year-out exam at the same time.

Retention goal	Optimal first review	Second review	Third review
1 week (exam)	1-2 days	3-4 days	Day 6
1 month	10-11 days	3 weeks	Day 28
1 year	3-4 weeks	8-12 weeks	6 months
Permanent	1 month	3 months	6+ months

A practical schedule built on the Cepeda et al. (2008) ratio: the longer you need to keep it, the wider the first gap.

This is why the 1-3-7-14-30 ladder on study blogs is a heuristic rather than a finding, and why the most-cited ladder of all, 1/3/7/21, has no paper trail whatsoever. One counterintuitive result deserves more attention than it gets: Karpicke and Roediger found that equally spaced reviews match or beat expanding ones for long-term retention. The intuition that gaps must keep growing holds mainly for short-term recall.

Why I Capped LearnClash’s Ladder at Three Stages

When I built LearnClash’s review system, the reference designs all pointed one way: an open-ended interval ladder in the Anki style, where every card carries its own ever-growing gap. I went the other way and shipped a state machine with exactly three stages. Learning, Known, Mastered, and nothing else.

The entire scheduler is two constants in our codebase: a three-value stage enum and a cooldown array that reads [7, 90]. A card in Learning comes back in 7 days, a card in Known comes back in 90, and a miss demotes it exactly one stage instead of resetting it to zero. Mastered does not even have an interval; we stamp those cards with a next-review date of December 31, 9999, because the review pool is a Firestore range query on that date (“everything due by now”), and a null or missing date would silently fall out of that query instead of meaning “never due”, so “never see this again” is literally implemented as an appointment nearly eight millennia out.

The trade was deliberate. FSRS fits about 21 parameters per user and will beat a three-stage machine on scheduling precision; I never disputed that. But LearnClash players already track one number that moves every game, their ELO rating, and asking them to also trust a second, opaque model felt like one number too many. When a player asks why a card came back today, I want the answer to be 7 or 90, not the output of a fitted curve.

That 1.7 percent looked alarming the first time I pulled the census. Then I did the arithmetic: a card cannot reach Mastered without a correct answer on both sides of a full 90-day gap, and at export time the system had been live for barely two of those cycles, so graduation was mathematically capped no matter how well people played. The full decision history, including the folklore schedule we refused to ship, is in why we threw out the 1/3/7/21 ladder, and the LearnClash SRS retention curve walks through how the checkpoints behave in production.

Active Recall and Spaced Repetition: Better Together

Spacing decides when you practice. The other half of the equation is what you do at each review, and the evidence there is just as lopsided.

Karpicke and Roediger (2008) had students learn word pairs, one group by restudying, the other by repeated recall tests. During the learning phase the two groups looked identical, which is exactly what makes rereading so seductive. A week later, the testing group retained 80 percent of the material. The restudy group kept 36.

The two groups tie during practice and separate at one week (Karpicke and Roediger, 2008).

The two techniques compound because they answer different questions. Active recall fixes what you do: retrieve, don’t reread. Spaced repetition fixes when: just before forgetting, not at random. Combined, every review is maximally effortful at the most useful moment.

Competition is harder to quantify honestly. Big retention multipliers get claimed for gamified review, and I do not trust most of them, because the sourcing rarely survives a second look. What competition demonstrably changes is whether you show up at all. Getting a question wrong in front of an opponent also stings in a way a solo flashcard miss does not, and my unscientific observation from our own player base is that those misses tend to come back correct at the 7-day check.

For the speed-and-tactics side, how to memorize fast covers memory palaces, mnemonic systems, and the tactics that pair well with spaced retrieval when you are up against a deadline.

Does Quizlet Have Spaced Repetition?

Quizlet’s free Learn mode spaces questions only inside a single study session and resets when you start over, with cross-session review reserved for its paid plans; Anki runs full cross-session SRS but is solo-only with a steep learning curve; LearnClash builds complete spaced repetition into competitive 1v1 quiz duels, so the scheduling runs while you play against a real opponent. The Kahoot vs Quizlet comparison covers the classroom side of the same question.

	Quizlet	Anki	LearnClash
SRS algorithm	Basic in-session (free); Memory Score (Plus)	SM-2 / FSRS (full)	3-stage SRS (full)
Cross-session scheduling	Free Learn: no; Plus adds Memory Score and Scheduled Review	Yes	Yes
Competition	No	No	1v1 ELO-ranked duels
Topic coverage	User-created decks	User-created decks	Any topic, matched to your level
Difficulty levels	No	Card-level	Easy, medium, hard (ELO-calibrated)
Price for full SRS	$35.99/yr (Quizlet Plus)	Free (desktop and Android), $24.99 (iOS)	Free (daily duels + SRS)

Quizlet adapts question order within a session based on what you miss. That makes it genuinely useful for cramming before a test, and much weaker for remembering anything in six months: nothing carries over once a free Learn session ends, and cross-session review sits behind Quizlet Plus’s Memory Score and Scheduled Review.

Anki is the gold standard for pure spaced repetition: Wozniak’s SM-2 or the newer FSRS, full cross-session scheduling, community decks on nearly every subject. It is also a solo grind with an interface that has barely moved since 2006, the eat-your-vegetables of learning apps. The dedication ceiling is real; in 2010, computer scientist Roger Craig used Anki with the J-Archive to set what was then the single-day Jeopardy winnings record, $77,000 (broken by James Holzhauer in 2019). For a head-to-head on pricing and medical school adoption, see Anki vs Quizlet; the wider field is in the Quizlet alternatives comparison, and there is a separate look at the licensed Jeopardy app.

LearnClash folds the same science into a game loop. Miss a question during a duel and it enters your review queue without ceremony; answer it correctly at two spaced checkpoints and it retires. There is no separate review session to schedule, which for most players is the difference between a system they admire and a system they use.

How LearnClash Runs Spaced Repetition While You Play

Two surfaces feed the same review pool. Solo mode is the deliberate one: short sessions built from the cards your personal schedule says are due. Duels are the ambient one, mixing your due questions in with fresh ones, and your opponent has no idea which is which.

The stages work the same everywhere:

• Learning: you just missed it, or it is new. Returns in 7 days.
• Known: right once. Returns in 90 days for confirmation.
• Mastered: right again after the 90-day gap. The card leaves the review pool.

A miss never resets a card to zero. It drops one stage, and the cycle continues from there.

Miss a question and the system handles the rest of the calendar.

In practice, the duels do most of the work: in LearnClash’s July 21, 2026 production export, about 93 percent of SRS-tracked first encounters happened inside duels rather than Solo mode, so most review scheduling starts as a side effect of someone accepting a challenge.

For the research on why quizzes beat rereading, see the testing effect. To slot spacing into a full routine with focus, exam, and sleep tactics, how to study effectively covers the other eight methods, and how the ELO rating system works explains the matchmaking that keeps duels close.

What I would tell a student setting this up tonight: pick the one subject that worries you most and run its material through a single test-yourself pass before bed, because tonight’s misses are the seed of the whole schedule. Then leave it alone; the gap is the active ingredient, and the itch to re-check tomorrow is the illusion of competence talking. When the material comes back in a week and you struggle, that struggle is the system working, not failing. And if your exam is five days away, cram anyway, take the short-term win, then run one spaced pass afterward for the version of you who meets this material again in a year. Start with a duel on any topic →