How many flashcards a day? The review math, simulated

The short answer

Decide how much time you can give flashcards every day for the long run, then add new cards at a pace that fits it a year from now, not just this week. At 10 seconds a review and a 90% retention target, a year in, each new card a day costs about 1.6 minutes of daily review:

In the first weeks the load is much lighter, so you can go faster for a while, for example before an exam, as long as you know the bill comes later. The rest of this article shows where these numbers come from.

How we simulated it

We used py-fsrs 6.3.2, the open-source implementation of the FSRS algorithm, with its default parameters and the settings FlashCardz uses: a 90% desired retention, learning steps of 1 and 10 minutes, a 10-minute relearning step, and no random fuzz on intervals. Then we followed 4,000 cards for two years:

• Each card is seen for the first time and rated Good: you've just read it, so you know it.
• Every later review is remembered with exactly the probability FSRS predicts for that moment, and rated Good if remembered, Again if not.
• Every review is done on the day it's due. Nobody misses a day.
• A review takes 10 seconds. New cards take longer than that, so budget extra time for them.

Because cards don't affect each other, a student adding N new cards a day has N times one card's reviews, summed over the cards' ages. As a check, the simulated students remembered 90.0% of their due reviews, exactly the target. Real students rate some cards Hard or Easy, miss days and make cards of different difficulty, so treat these as typical figures, not promises.

Reviews a day, by how many new cards you add

Average reviews a day in the week ending on each day, aiming for 90% recall:

A simple rule falls out of it: daily reviews are about three times your new cards a day after the first week, five times after a month, seven times after three months, ten times after a year and twelve to thirteen times after two. The numbers scale exactly with the pace, so for 15 new cards a day, take the 10 row and add half.

In time, at 10 seconds a review, 20 new cards a day comes to about 16 minutes of reviews after a month and 33 minutes after a year. 50 new cards a day, a pace some students try before exams, comes to about 40 minutes after a month and 82 minutes a day after a year.

Where one card's reviews go

Following a single card, aiming for 90% recall, the reviews add up like this:

Half of a card's first-year reviews come in its first month, when its memory is still fragile, which is why a burst of new cards hits hardest three or four weeks later. After that, each card asks for less and less: under three reviews in the whole second year. The cards you keep forgetting take most of those reviews; the ones you know well hardly come up.

The retention trade-off

FSRS lets you choose a desired retention: the chance you'll still recall a card when it comes up. The higher you aim, the shorter the gaps and the more reviews. With 20 new cards a day:

Going from 90% to 95% halves how often you forget a card when it comes up, from one time in ten to one in twenty, but costs about 45% more reviews a year in. Dropping to 80% saves about a quarter of the reviews, but you'll miss twice as many cards, which also means more relearning and more frustration. That's why 90% is FSRS's default and the target FlashCardz uses. In the first month, 80% and 85% cost about the same, because the day you learn a card and the first reviews after it dominate early on.

Why the pile keeps growing

The daily load climbs fast for the first month or two, then more and more slowly, but it never quite stops: cards from a year ago still come back now and then, and you keep adding new ones. From the end of the first year to the end of the second, reviews grew by about 27% in the simulation, with the same number of new cards each day.

Two consequences. Choose a pace you can sustain for the whole course, not only the first exciting fortnight. And when a course ends, decide what to keep: material that later courses build on, like most of medicine and law, is cheap to maintain, while cards for an exam you've passed and will never need again can be set aside.

Work backwards from your exam

1. Estimate how many cards the exam needs. A dense lecture often gives 15 to 40 good cards; multiply by the lectures.
2. Count the days until the exam, and set aside the last 7 to 14 days for review and practice questions only.
3. Divide: cards needed ÷ days left for new cards = new cards a day.
4. Check the load: about five times that pace in reviews a day after a month, at 10 seconds each. If it's more time than you have, cut the cards for minor topics, not the reviews.

Example: 1,200 cards and eight weeks to go. Keep the last 10 days for review, which leaves 46 days, so about 26 new cards a day. A month in, that's around 130 reviews, roughly 20 to 25 minutes, plus the time to learn the new cards. Manageable, if you start now; impossible if you start two weeks before. See how to study for exams and cramming vs spaced repetition.

Signs you're adding too many

• Your reviews take longer every week and you've started skipping days.
• You're failing more than about one card in five that comes up, when you're aiming for 90%.
• You recognise the cards but can't explain the answers: you're memorising cards, not ideas.
• New cards feel like a chore you rush through.

If so, stop adding new cards for a few days, clear the reviews that are due, and restart at a lower pace. Don't reset your deck or delete your history: the scheduler needs it, and most of what you learned is still there.

Fewer, better cards beat more cards

The cheapest review is the one you never need. A card that tests one clear idea is reviewed quickly and rarely forgotten; a vague or overloaded card is forgotten again and again, and every lapse brings it back for more reviews. Before adding a card, ask whether the idea is worth ten reviews this year. Our guide to making flashcards that work covers how.

FlashCardz plans each day for you: the reviews that are due, a few new cards, and the course with the nearest exam first. It writes cards from your own lectures for the ideas that matter most, ranked, rather than for every sentence, and it schedules all of them with FSRS on every plan. See how it works.