The scheduling
The right word, on the day you were about to forget it
Reviewing something you still remember is wasted time. Reviewing it after you have forgotten is wasted learning. FSRS is the part of Rayaad that aims for the narrow window between the two — and it does it per word, per person, without asking you to set anything.
- Two answers. No four-point scale to second-guess.
- No intervals to choose and no parameters to optimise.
- The same schedule on every device, replayed from your full history.
Didn’t know
in about a minute
It stays in this session, a few cards later.
Knew it
tomorrow
It leaves the session, and the gap grows from there.
Know it a second time and it comes back in 3 days.
Every interval here is computed by the app’s own scheduler.
What it does, in one picture
Each right answer roughly doubles the gap
These are Rayaad's real intervals for a card answered correctly every time, produced by the scheduler itself. Answer eight times over the course of a year and you have spent about four minutes on a word you now know.
Retrievability
How likely you are to recall this card right now. It falls as time passes, and the scheduler puts the card in front of you as it approaches the target.
Stability
How slowly that fall happens — the durability of the memory. Every correct answer increases it, which is why the gaps grow.
Difficulty
How stubborn this particular card is for you. Two words reviewed the same number of times do not get the same interval if one of them keeps catching you out.
Getting it wrong
A lapse is a correction, not a punishment
Answer "didn't know" and the card comes back inside the same session — about a minute later, after a few other cards have gone past so you are recalling it rather than reading it off the screen. What it does not do is throw away everything you had built up.
Measured on a card four correct answers in: its durability drops from about48 days to3 days — reduced, not reset. It returns about a minute, and once you get it right it is scheduled tomorrow and starts climbing again.
- in 19 daysFour right answers in. The card is weeks away.
- You don't know itSwipe left. It returns in about a minute, still in this session.
- tomorrowYou get it. Not back to day one — back to where it can climb.
Why not just fixed intervals
1 day, 3 days, a week, a month — for every word, for everyone
A fixed ladder is easy to explain and wrong in two directions at once. It shows you words you already know solidly, because the ladder does not know that; and it shows you words you have half-forgotten too late, because the ladder does not know that either. The cost is not theoretical — it is the reviews you sit through for nothing and the words that slip away between steps.
A fixed schedule
- The same gaps for a word you find trivial and one you keep missing.
- No memory of how close you were to forgetting.
- Miss a week and the ladder has no answer for what to do next.
FSRS
- Per-card durability and difficulty, updated on every answer.
- A target recall probability rather than a step number.
- A gap of 11 months for a word that has earned it.
Exactly what Rayaad ships
The settings, published rather than described
Read straight out of the configuration file the app and the server both schedule against. If it changes, this table changes with it.
| Setting | Value | What it means |
|---|---|---|
| Algorithm | FSRS-6 | The current generation of FSRS, via the ts-fsrs@5.4.1 implementation. |
| Target retention | 95% | The recall probability the scheduler aims for at the moment a card comes back. |
| Learning and relearning steps | 1m / 1m | One short step, then straight into real intervals. No multi-day drilling ladder. |
| Interval fuzz | Off | Two identical histories get identical schedules, which is what makes a schedule reproducible and auditable. |
| Maximum interval | 36,500 days | Effectively no ceiling. A word you have truly learned stops costing you anything. |
| In-session retry floor | 90s, 5 cards | A missed card cannot come straight back — other cards go past first, so you recall it rather than read it. |
| Scheduler version | fsrs-6-ts-fsrs-5.4.1-rayaad-v2 | Stored with every review, so a past schedule stays explainable after a parameter change. |
Reviews are append-only and a card's state is recomputed by replaying its whole history on the server, not by trusting a running total on a device. That is what makes the same account agree across the web, a phone and a browser extension.
Spaced repetition, in the plainest terms
Everything you learn starts fading immediately. The rate of the fade is not constant: each time you successfully pull something back out of memory, it fades more slowly the next time. Spaced repetition is the practice of arranging your reviews to land near the point of forgetting, because that is where a review does the most work per minute spent.
The hard part has never been the idea — it is knowing when that point is, for each item, for you. Doing it by hand means keeping a schedule for hundreds of words. Doing it with a fixed ladder means guessing the same answer for every word. FSRS does it by fitting a model of forgetting to the answers you actually give, which is why the intervals above are not round numbers.
Why vocabulary is the ideal case
Vocabulary is made of many small, independent items, each of which you either retrieve or do not. That is precisely the shape spaced repetition handles best — and it is why a vocabulary tool can be simpler than a general study tool without being weaker. There is no need for cloze deletions, image occlusion or a note-type editor to schedule a word well.
How much reviewing is this, actually
Less than people expect, because the whole point is that successful reviews get rarer. A mature deck settles into a small daily number that depends on how much you are adding rather than on how much you have learned. Rayaad shows you what is due before you start, so the size of a session is never a surprise, and there is no daily review cap on any plan — capping reviews in a spaced-repetition app would break the one thing it is for.
Next: what a review session actually feels like, orhow this compares with Anki's implementation.
FSRS questions
What does FSRS stand for?
Free Spaced Repetition Scheduler. It is an open algorithm for deciding when a card should next be reviewed, developed in the open-source spaced-repetition community and now used by several serious review tools.
How is FSRS different from the older SM-2 scheduler?
SM-2 keeps one number per card — an "ease factor" — and multiplies the interval by it. FSRS models memory with three quantities instead: how likely you are to recall the card right now, how durable that memory is, and how hard this particular card is for you. Because it separates durability from difficulty, a card you find hard and a card you have simply not seen in a while are treated differently rather than being collapsed into one multiplier.
Do I have to configure or optimise anything?
No. Rayaad ships one parameter set for everyone and there is no settings screen for it. That is a deliberate trade: a tool that lets you optimise parameters against your own history can fit you better, and it also gives you something to get wrong and to worry about. Rayaad chose "already tuned, nothing to touch", and the exact parameter set is published as fsrs-6-ts-fsrs-5.4.1-rayaad-v2.
Why only two answers instead of four?
Because the middle two are the ones people disagree with themselves about. FSRS works correctly when only "again" and "good" are used — that is a supported way to use it, not a degraded one — and removing the choice removes a decision from every single card, several hundred times a week.
What retention does Rayaad aim for?
95%. That is the probability the scheduler is targeting that you will recall a card at the moment it comes back. Aiming higher means shorter gaps and more reviews for the same material; aiming lower means fewer reviews and more forgetting.
What happens if I stop for a month?
Nothing is lost and nothing is punished. Cards that came due while you were away are simply due; you answer them, and the ones you have genuinely forgotten go back to short intervals while the ones that stuck keep their long ones. There is no streak to break and no penalty for the gap.
Can I see why a card was scheduled the way it was?
Yes. Every review is recorded permanently with the scheduler version that produced it, and the server recomputes a card’s state by replaying its entire history rather than trusting a running total. That is what makes a schedule explainable after the fact instead of merely plausible.
See it on your own words
The demo at the top of this page runs the real scheduler. The app runs the same one on a deck that is yours.