The right moment. For every single one of your cards.
Qardbox schedules your reviews with FSRS-6, a method that estimates for every single card when you are likely to forget it. And it then sets the next date shortly before that, exactly in the window where recall still feels like an effort and still works.
Why shortly before forgetting
Review too early and the material is still sitting there anyway; review too late and it is long gone. In between lies a window of a few days, and that window is exactly what FSRS hits for you, card by card.
For that, the scheduler keeps estimating how likely you are to still remember a card. That probability falls with every day, and it falls at a different rate for every card. Qardbox sets the next date on the day it has dropped to 90 per cent. The picture below shows, on a single card, what that produces over five reviews.
That is why your intervals grow all by themselves over time. A card you knew confidently three times will not come round again for weeks, and one you keep getting stuck on you will see the very next morning.
known every timeforgotten the third time
Calculated with the same formulas that run in the app, and checked against its test data. The bars show one single new card over five reviews. Example values, no collected usage data.
The intervals do not grow evenly, they grow by a multiple each time. Anyone who knows a card five times in a row will not see it again for a year and a half. A single lapse at the third review throws it back from 46 days to 3, and from there it grows again.
That is not a punishment, it is the consequence of the same calculation. If you have forgotten a card, your memory of it was weaker than estimated, and the scheduler lowers the stability accordingly. After that it climbs again faster than it did the first time.
Stability is the interval
The 90 per cent is not a rounded rule of thumb, it is the hinge of the whole calculation. Put the day that corresponds to a card’s stability into the formula, and exactly 0.9 comes out. A card’s stability is therefore nothing other than the number of days up to precisely that point, and both terms basically name the same thing with two different words.
From that follows the simplest sentence anyone can say about this scheduler. The interval until the next review is the stability in days, rounded to whole days. If a card stands at 21, you will see it again in exactly 21 days.
This 90 per cent is fixed in Qardbox and cannot be adjusted. A lower value would stretch your intervals and hand you fewer reviews, and leave more gaps in return. The app takes that trade-off off your hands completely, and you never have to concern yourself with it.
Where FSRS comes from and who is behind it
FSRS stands for Free Spaced Repetition Scheduler and is an openly published, data-driven method of scheduling reviews that gets by without a fixed box scheme. Qardbox implements FSRS-6 independently, following the specification. And no source code from the reference implementation was taken over in the process.
Specification and reference implementations: Jarrett Ye and others, open-spaced-repetition, licences MIT and Apache 2.0.
What the learning principles rest on
They rest on well-documented memory research: the spacing effect says you should learn spread out rather than in one block, and the testing effect says you should retrieve rather than simply reread. And desirable difficulties mean that a little resistance during retrieval helps you.
The first evidence for it is old. Hermann Ebbinghaus measured on himself in 1885 how quickly what he had learned faded again, and Jaap Murre and Joeri Dros repeated and confirmed his experiment in 2015. What Ebbinghaus found, however, is the saving on relearning and not the share you still know.
Jaap M. J. Murre and Joeri Dros: Replication and Analysis of Ebbinghaus’ Forgetting Curve. PLOS ONE 10(7), 2015.
Qardbox derives no promise from that. How well you learn depends on your material and on your regularity. And Qardbox concerns itself solely with the timing.
For everyone who would rather know exactly
There are learning programs that tell you they schedule cleverly and leave open what they schedule by. Here you find everything you need to check the arithmetic. The method is published, the formula is on this page, and the numbers in the picture above were calculated with it.
The principles behind it do not come from a product idea but from memory research, and the oldest evidence for it is over a hundred years old. What Qardbox derives from that is explicitly not a promise of results, because how well you learn in the end depends on your own material and on your regularity.
The terms that come up in the app
These words keep coming up in the app. And each of them has its own anchor address here, which you can link to.
- Due
- A card is due when its scheduled review date has been reached. Due cards come up first.
- New
- Cards you have never learned yet. They do not have a scheduled date yet. As soon as you answer them for the first time, FSRS takes them into the schedule.
- Review
- Every card you answer counts as one review. That is the same as one learned card.
- Stability
- The number of days after which your memory of a card has fallen to 90 per cent. It is therefore at the same time the interval until the next review, rounded to whole days. A new card has a stability of a few days. The more often you answer it correctly, the longer it lasts, and if you forget it, it falls back sharply. In the app’s insights it appears under the word Durability.
- Difficulty
- How hard a particular card is for you in the long run. It changes slowly and influences how strongly the stability grows on correct answers.
- Retention rate
- The share of your answers that were correct, averaged over seven days at a time. Around 90 per cent is the algorithm’s target.
- Mastered
- A card counts as mastered when it sits so firmly that the next date is around a month away or further and you have answered it correctly several times.
When Qardbox readjusts the scheduling
From a certain amount of experience onwards, Qardbox replaces the standard values with your own. The first milestone is at 512 ratings a day apart, then follow 1,024, then 2,048 and from there always the next doubling. And the gaps grow on purpose: the more data is already in, the less a single run changes.
The calculation happens on your device, and Qardbox only adopts your own values if they end up fitting your learning behaviour better than the tested standard values. And you can see the history yourself under Settings, Learn, Personalized study planning.
Ratings a day apart
1,024
Milestone reached
Distinct cards behind them
312
Example values. These two numbers are in the app under Settings, Learn, Personalized study planning, Your history.
Why only ratings a day apart count
If you draw the same card twice on the same day, only one of them counts. What you still know ten minutes later says little about whether you will still know it in two weeks. And for the scheduling only the gap across days counts, because that is where it first shows what has really stuck.