Spaced repetition: the schedule that beats cramming (on paper, in a Leitner box, and what an algorithm's "desired retention 0.9" means in days)

Spaced repetition is reviewing an idea on a schedule of growing gaps — a day, then three, then a week, then two — so that each review lands just before you would have forgotten it. The same hours spread over days produce far better recall than the same hours in one sitting, across hundreds of studies. You can run it on paper with dates, in a box of index cards, or with software that computes the next date from how you did. All three are below, with the dates worked out.

What spaced repetition is

Two ideas are folded into the name. The first is spacing: a gap between one review of an idea and the next, instead of ten reviews in one evening. The second is that the gaps grow: a day, then three, then a week, then two weeks, then a month. Each review is a retrieval, not a reread. You look at the question, produce the answer with the back of the card hidden, then check. Each review is a retrieval, and the active recall guide is where the "produce, then check" rule is spelled out; this guide is about when. Spaced practice is ranked second in the evidence among the ten common study methods, right behind practice testing, and the two are meant to be used together.

The evidence

Cepeda and colleagues (2006) reviewed hundreds of experiments in which the same material was studied either in one block or spread over gaps. Spaced practice beat massed practice in the large majority of comparisons, across word lists, facts and texts (Cepeda et al. 2006). The review that rated ten study methods gave distributed practice its highest rating, alongside practice testing (Dunlosky et al. 2013).

How big should the gap be? Cepeda and colleagues (2008) tested gaps from minutes to months against tests held a week to a year later. The best gap grew with the time until the test: about a day's gap for a test a week away, and one to two weeks for a test a month or two away. The paper's own picture is a ridgeline: for any test date there is a gap that is too short, a gap that is too long, and a best gap between them, and the best gap is a fraction of the time to the test that gets smaller as the test gets further off (Cepeda et al. 2008). The rule of thumb that follows: for a test next week, review again tomorrow and then in a few days; for a test in two months, a week between reviews is about right.

Kornell (2009) ran the comparison students actually face. Learners studied flashcards either as one large stack reviewed on several days (spaced) or as small stacks each crammed in one go. Spacing beat cramming for about 90 per cent of the learners. Asked which had worked better, most of them, about 72 per cent, said cramming had (Kornell 2009). That gap between what worked and what felt like it worked is why students keep cramming, and the guide on cramming, and what it costs goes through what it buys and what it loses.

The paper schedule, with dates

The 1–3–7–14–30 pattern is a convention. The research does not hand down those exact numbers; it says gaps should grow and should scale with the test date, and this pattern does both while staying easy to run with a pen. Write the next review date on the card each time you get it right. Here it is for a card made on Monday 9 November 2026.

Review	Gap since the last	Date
Made	—	Monday 9 November
1st review	+1 day	Tuesday 10 November
2nd review	+3 days	Friday 13 November
3rd review	+7 days	Friday 20 November
4th review	+14 days	Friday 4 December
5th review	+30 days	Sunday 3 January 2027

One rule covers misses: a card you get wrong goes back to +1, whatever stage it was at, and climbs again from there. Nothing else changes. If you miss a day, do the cards that were due and carry on from today's date; do not try to do the week you missed in one sitting.

Chapter 2 of the free playbook, "Space it out", has this table as a page to print and a column to write the dates in. The dates are the only bookkeeping the method needs, and the point of writing them on the card is that you then do not have to remember the schedule — you only have to look at today's date.

The Leitner box

The Leitner box is the same schedule run with compartments instead of dates. It comes from Sebastian Leitner's 1972 book So lernt man lernen ("How to learn to learn"), and it needs a shoebox and five dividers.

1. Label five compartments 1 to 5. Every new card starts in compartment 1.
2. Review compartment 1 every day. A card you get right moves forward to compartment 2; a card you miss stays in 1.
3. Review compartment 2 every two or three days. Right moves it to 3; wrong sends it back to 1.
4. Review compartment 3 once a week, compartment 4 every two weeks, compartment 5 once a month. The same rule each time: right moves forward, wrong goes back to compartment 1.
5. A card that is answered right in compartment 5 is done. Take it out, or leave it for a monthly check before the exam.

Write the review days on the dividers (compartment 2: Tuesday and Friday; compartment 3: Sunday) so the box tells you what is due. A wrong answer always goes back to the start, not one step back; that is what keeps a weak card coming round often.

What the software does (and what "desired retention 0.9" means)

A spaced-repetition app does the dates for you, and the better ones do more than count days. Behind the schedule is an algorithm that models each card's memory separately and picks the next date from how you did. The one worth understanding is FSRS, which is an open algorithm, not a product, and is used by several flashcard apps.

FSRS estimates, for each card, a number it calls stability: the number of days until your probability of recalling the card is predicted to drop to 90 per cent. Set the app's desired retention to 0.9 and the schedule falls out directly. A card with a stability of 7 days comes back in 7 days. A card with a stability of 30 days comes back in 30. When you answer a card right, its stability grows, so the next gap is longer; when you miss it, the stability shrinks and the card comes back soon. Lower the desired retention to 0.8 and every gap gets longer: fewer reviews, and more cards forgotten by the time they come due. Raise it to 0.95 and the gaps shorten: more reviews, less forgetting. The number is a trade between your time and your memory, and 0.9 is the usual middle.

What the software cannot do is as important. It cannot write the card: a card that asks a bad question comes due on schedule and teaches nothing each time. And it cannot make you attempt the answer before you flip. Every schedule above assumes that each review is a retrieval. The streak counter is not a measure of learning; the dates are only as good as what you do when the card comes up. The AI-for-studying section has a guide on how to choose a study app that puts those two tests first.

Worked example: twenty cards, three weeks, one chemistry unit

SCH4U, the idea called organic functional groups and IUPAC nomenclature. Twenty cards: ten of the form functional group → suffix, ten of the form structure → name. Two of them, with their answers:

• "Name CH3CH2CH2OH." Three carbons, so propan-; an alcohol, so -ol; the OH on carbon 1: propan-1-ol.
• "What suffix marks a ketone?" -one (as in propanone).

Here is the paper schedule applied to all twenty, made on Monday 9 November, with a realistic run of misses: six on the first night, and one or two on the heavier days after.

Day	Cards due	Of which missed	What happens to them
Mon 9 Nov	20 (made)	6	14 right → Thu 12 Nov; 6 wrong → Tue 10 Nov
Tue 10 Nov	6	1	5 right → Fri 13 Nov; 1 wrong → Wed 11 Nov
Wed 11 Nov	1	0	→ Sat 14 Nov
Thu 12 Nov	14	2	12 right → Thu 19 Nov; 2 wrong → Fri 13 Nov
Fri 13 Nov	7	1	4 → Fri 20 Nov; 2 (restarted) → Mon 16 Nov; 1 wrong → Sat 14 Nov
Sat 14 Nov	2	0	1 → Sat 21 Nov; 1 → Tue 17 Nov
Mon 16 Nov	2	0	→ Mon 23 Nov
Tue 17 Nov	1	0	→ Tue 24 Nov
Thu 19 Nov	12	0	→ Thu 3 Dec
Fri 20 Nov	4	1	3 → Fri 4 Dec; 1 wrong → Sat 21 Nov
Sat 21 Nov	2	0	1 → Sat 5 Dec; 1 → Tue 24 Nov
Mon 23 Nov	2	0	→ Mon 7 Dec
Tue 24 Nov	2	0	→ Tue 1 Dec and Tue 8 Dec

Three weeks, 75 reviews in all, on 13 of the 21 days. After the first night, the heaviest day is 14 cards and most review days are under 8; eight days have nothing due. That is the answer to the objection students raise about spacing — that it means doing everything every day. It means the opposite. The gaps grow, so the daily load stays small, and the cards you keep missing are the ones you see most.

Notice, too, what happened to the card missed on Friday 20 November: it went back to +1 and will climb again. By the exam it will have been retrieved more times than any card you got right the first night. That is the schedule working as intended. The cards themselves matter as much as the dates; the guide on cards that force recall is about writing a question side that cannot be answered by recognising the shape of the answer.

Spaced repetition for problems, not just facts

Nothing here is limited to facts. A maths problem you got wrong goes into the same schedule: write the question on a card, and when the card comes due, redo the problem from a blank page before you look at the solution. The guide on how a logged problem comes back as a redo has the marking rule. A derivation works the same way: the card says "derive the quadratic formula", and the review is the derivation. For a whole course, the dates live on a weekly grid rather than on cards, and the weekly grid where the dates live shows one filled in.

The mistakes

• Reviewing by rereading the answer side. The schedule is right and the review is wasted. Hide the answer, produce it, then check.
• Every card every day. That is not spacing; it is a daily cram, and it stops working the week you cannot face the pile.
• Abandoning the box after a missed week. Restart from today's dates, do what is due, and move on. Do not binge the backlog.
• Cards that are keyword lists, not questions. "Alkanes — single bonds — -ane" on the front tests nothing. One question, one answer.
• Confusing a flashcard app's streak with learning. A streak counts days opened. The test counts what you can produce.

Summary box

• Review on growing gaps: +1, +3, +7, +14, +30 days, written on the card; a miss goes back to +1.
• Or run a Leitner box: five compartments, right moves forward, wrong goes back to the first.
• Software does the dates by estimating each card's memory; at a desired retention of 0.9, a card comes due on the day you are predicted to be at 90 per cent.
• The one rule under all of it: attempt before you look. A schedule of rereads is a schedule of nothing.
• The daily load stays small because the gaps grow; the cards you miss are the ones you see most.