Spaced repetition: how it works, from Leitner to FSRS

Spaced repetition is a way of organizing your reviews. Instead of rereading the whole course the night before the exam, you review each concept several times, leaving an interval between two reviews that grows with each success: the next day, then a few days later, then a week or two, then a month.

The principle fits in one sentence: each successful recall makes the memory stronger, so you can wait longer before the next one. Conversely, a forgotten concept comes back quickly. Your study time is thus concentrated on what you have not mastered, not on what you already know.

In practice, it is almost always applied with flashcards: a question on one side, the answer on the other. You try to answer from memory, flip the card, then say honestly whether you knew it. That judgment sets the date of the next review. Spaced repetition is therefore inseparable from active recall: you don't reread, you test yourself.

In 1885, the German psychologist Hermann Ebbinghaus published a series of experiments he ran on himself. He learned lists of nonsense syllables, then measured how long it took him to relearn them after varying delays. From this he drew the forgetting curve: we forget very quickly in the hours and days after learning, then the loss slows down.

He also observed the other side, often overlooked: relearning goes faster than the first time, and each review flattens the next curve. A memory reviewed several times fades more and more slowly. This is exactly what spaced repetition exploits.

For a student, the consequence is concrete. A biochemistry lecture understood on Monday will be largely lost by the end of the week if you don't come back to it. A short review the next day, then a few days later, costs far less than relearning everything a month later.

The spacing effect is a very robust finding in cognitive psychology: for the same amount of study time, spreading your sessions out over time produces better long-term retention than grouping them into one long session. The meta-analysis by Cepeda et al. (2006), which brings together hundreds of experiments on distributed practice, confirms it under a wide range of conditions.

Cramming the night before feels effective because everything is fresh the next morning. But what was learned in one block is also forgotten in one block. For subjects that build up from one year to the next, such as anatomy, contract law or prepa mathematics, this difference ends up weighing heavily.

The methods have evolved from cardboard boxes to algorithms. Each keeps the same idea, but decides the intervals in an increasingly precise way.

The Leitner system remains an excellent way to understand the principle, and it works with simple index cards. SM-2 was for decades the core of many programs, including Anki. FSRS, a free algorithm developed notably by Jarrett Ye, is today the most accurate approach available in open software, and Anki offers it too.

FSRS describes each memory with three quantities:

• Stability: the time, in days, after which the probability of remembering the card drops to a reference level. The greater it is, the more durable the memory.
• Difficulty: how much this particular card resists. A difficult card sees its stability increase more slowly after each success.
• Retrievability: the probability, at a given moment, that you will find the answer. It decreases with the time elapsed since the last review, and all the more slowly as stability is high.

At each review, your answer (again, hard, good, easy) updates stability and difficulty. A success obtained when retrievability was low increases stability sharply: it is the difficult but successful recall that consolidates the most. A failure makes stability drop, and the card comes back soon.

The algorithm then calculates the date on which retrievability will fall to the retention you are aiming for, for example 90%. That is the day the card reappears. You see it neither too early, which would waste time, nor too late, which would force you to relearn it.

Target retention, or desired retention, is the recall probability at which you accept to review a card. It is the main FSRS setting, and it has a direct cost.

• Higher (for example 95%): you forget less, but intervals shorten and the number of daily reviews rises sharply.
• Lower (for example 80%): fewer reviews, but more cards forgotten at each pass.
• Around 90%: the most common default setting, a reasonable compromise for most subjects.

Raise the target for a decisive deck as a competitive exam approaches, lower it for general-knowledge material. The retention gain above 90% is paid for with many extra reviews: don't aim for 99% on principle.

• Letting overdue cards pile up. A few days without reviewing, and the stack becomes discouraging. A short session every day is better than an hour of catching up on Sunday. If you are behind, pause new cards until you have cleared the backlog.
• Adding too many new cards per day. Each new card generates reviews for weeks. Fifty new cards today means hundreds of reviews to come. Set a cap you can keep up for the whole semester.
• Making cards that are too long. A card that asks you to recite a paragraph or a list of eight items is almost always graded "more or less". The algorithm can't learn anything from it. Split it up: one idea per card, as explained in how to make good flashcards.
• Cheating on your grading. Clicking "good" when you hesitated pushes the card too far out. The algorithm is only as good as your honest answers.
• Memorizing without understanding. Spaced repetition maintains what has been understood; it does not replace the course.

Mnemesia schedules cards with FSRS. The desired retention is adjustable, so you choose for yourself between fewer reviews and less forgetting. Cards can be front and back, cloze deletion, image occlusion or formulas, and you can bring in your Anki decks (.apkg) or your lists copied from Quizlet (CSV, TSV): see Anki alternative.

As exams approach, the multi-deck “Review for exams” mode gathers several decks into a single session, to review a whole subject in one go. It works together with the study planner, which spreads the workload over the remaining days.

The app is free, with no account and no ads; it works offline and your cards stay on your device. It is available on Mac and iPad (App Store), Windows, Linux and Android tablets: download Mnemesia.