You finished a lesson on computed properties last week and passed its quiz. Today one of those questions may be back in your daily mix, labelled "Due for review". Nobody chose that day by hand. A scheduling algorithm set it from how you answered, and once you know how it works, you can let it decide when you review and keep each session short.
The forgetting curve in one paragraph
In the 1880s Hermann Ebbinghaus memorised lists of nonsense syllables and tested himself at intervals, and found that what he could recall fell off steeply at first and then more slowly. The shape he drew is called the forgetting curve. The steep start is the part that matters for learners: much of what you can recall at the end of a lesson fades within a day or two if nothing brings it back. Rereading slows the loss a little. Being made to recall it slows the loss a lot, and that is the whole basis of what follows.
Review just before you forget
Every successful recall does two things. It brings the idea back to full strength, and it makes the next decline slower. So the curve after a second meeting is flatter than after the first, and the curve after a third is flatter still. The question is when each meeting should happen.
Too early, and recall is effortless, which teaches you nothing new; you knew it already. Too late, and the memory is gone, so you are relearning rather than reviewing. The sweet spot is just before you would have forgotten: late enough that recall takes real effort, early enough that it still succeeds most of the time.
Each review at that point earns a longer gap before the next one. A question you answered after one day comes back after a few days, then a week or two, then a month, then several months. That stretching is called spaced repetition. It is why a small number of well-timed reviews can hold hundreds of ideas in memory, while ideas crammed into a single evening fade within days.
What FSRS does with each answer
The school uses an open-source algorithm called FSRS, the Free Spaced Repetition Scheduler, to decide the dates. Its default settings were fitted to millions of real reviews from flashcard users. Anki, a widely used flashcard app, has offered it as an option since version 23.10 in October 2023.
FSRS describes each item you are learning with three numbers:
- Retrievability is the probability that you can recall the item right now. It falls as time passes since the last review.
- Stability is how slowly retrievability falls, measured as the number of days until it drops to ninety percent.
- Difficulty is how hard the item has proved to be for you, on a scale from one to ten. The higher it is, the less each success raises stability.
On the school, every quiz question you answer becomes one card with those three numbers. We never ask you to rate your own answer, because the quiz already knows whether you were right, so there are only two outcomes:
- A right answer is rated "good". Stability goes up, and the next due date moves further out. Stability rises more when the review came late and recall was less likely than when it came early.
- A wrong answer is rated "again". Stability drops sharply, and the card is due again within minutes, so it sits near the top of the next daily mix you open.
A brand-new card is handled differently at first: after your first answer, right or wrong, it is due again within minutes. After that, each right answer sets the next date by solving the forgetting curve for the moment when retrievability will fall to the school's target of ninety percent. With that target the gap in days is about the card's stability, so a card you have answered correctly four times waits far longer than one you just missed.
The cards live in your browser, next to the rest of your progress. There are no accounts, and nothing about your answers is sent anywhere unless you have chosen a hosted AI provider in the settings.
Where due questions show up
The schedule appears in three places:
- The progress page shows how many questions are due today and groups the upcoming reviews by day, for the next few days that have any.
- The practice page marks a finished lesson's set as due when the lesson is slipping: one of its questions is due and you missed something in its last quiz, or most of your last few answers on it were wrong.
- The daily mix on the practice page is built from the cards.
That mix is five questions, picked by scoring every question from lessons you have touched. A question whose card is due gets the largest boost. One you got wrong the last time you saw it gets the next largest. Questions from lessons where your recent answers have gone badly get a little more, and lessons you are weaker in overall count for a little as well. The five highest scores are taken, with no more than two from the same lesson, so one bad day on slots does not fill your whole mix with slot questions.
Each question shows its reason above the prompt: "Due for review", "Missed last time", "From a lesson that needs review", or simply "Keeps it fresh" when nothing is urgent and the pick is there to stop an older idea from fading.
If you have turned on the optional AI instructor, it can choose which questions to show you and offer hints when you are stuck. It picks from the same reviewed bank and it never writes questions. The scheduling itself is not something it changes.
How to study with the schedule
- Do the due reviews before you start a new lesson. They are the most valuable minutes of the session, because each one lands at the point where recall is hard and still possible, and they take less energy than new material. Three or four due questions take two minutes. Leave them until after the lesson and they become the thing you skip.
- Keep sessions short and frequent. The algorithm assumes you will be there when a card is due. A twenty-minute session most days keeps the queue small and the gaps growing. A long session once a week lets dozens of cards fall past their date. When you come back, many of them will be misses, which pulls their stability back down and puts you near the start of the curve for those ideas.
- Do not cram. Answering the same question five times in one sitting feels productive and barely moves its stability, because the later repetitions happen while retrievability is still near one hundred percent and there is nothing for the review to strengthen. If you got a question wrong, read the explanation, go back to the one paragraph in the lesson it points at, and move on. The schedule has already booked the card to come back, and that return is where it will stick.
- Answer from memory, not from the lesson. Scrolling back to check before you commit turns a recall into a recognition, which does far less for the memory. The quiz still counts it as right, so the card is pushed out as if you had remembered it. Guess, then read the explanation, whichever way it went.
When the gaps feel wrong
Two feelings are worth expecting. The first is that a question comes back when you are sure you still know it. Answer it anyway; if you are right, the gap after this one will be much longer, and the twenty seconds were the cheapest insurance you will buy today.
The second is that a question comes back and you have forgotten it, even though the gap seemed reasonable. That is not the system failing. The target is a ninety percent chance of recall, not certainty, so even when reviews come on time, about one in ten is expected to be a miss. A miss followed by a reread still does its job: the card comes back soon, and its gaps start growing again from there.
Missed days do not break anything. Cards that fell due while you were away are simply due now, counted as today on the progress page. Come back, do the reviews only, and skip new lessons until the queue is clear. The schedule picks up where it left off.