Updated August 8, 2026

Spaced repetition retrievability, explained

Retrievability is the probability that you could recall a card right now, if someone put it in front of you this second. It is a number between 0 and 1, it falls as time passes since your last review, and it snaps back near 100% each time you successfully recall the card. In FSRS, the scheduling algorithm used by Moji - Learn Japanese Daily for iPhone, retrievability is the quantity the whole scheduler is built around: a card is scheduled so that when it comes due, its retrievability has fallen to your target retention, which in Moji is 90%. That is the short answer to “when should I review Japanese vocab”: not on a fixed ladder of days, but at the moment your predicted recall crosses 90%.

Retrievability, stability, and difficulty

FSRS tracks three numbers per card, and they are easy to confuse.

Stability is how durable the memory is, measured in days. It is the slow-moving one, and it grows every time you recall the card.

Retrievability is how likely you are to recall the card at this instant. It is the fast-moving one, and it depends on stability plus how long it has been since you last saw the card.

Difficulty is how stubborn this particular card is for you, on a 1 to 10 scale. It modulates how much stability grows per review.

The relationship is worth internalizing: stability sets the shape of the decay, retrievability is your position along it. A card with 60 days of stability and one with 2 days both start at nearly 100% right after a review. Two weeks later the first is essentially untouched and the second is long gone.

The forgetting curve Moji uses

Moji uses the official FSRS-6 formulas with all 21 parameters (w0 through w20) at the FSRS-6 defaults. The forgetting curve is a power function, not the exponential decay most people picture:

R = (1 + F · t / S) ^ D

where t is days elapsed since the last review, S is stability, D is the decay exponent (the negative of w20, so -0.1542 at defaults), and F is a constant chosen so that the curve passes through exactly 90% when t equals S. That last detail is the hinge of the whole system: stability is defined so that “one stability” of elapsed time means “down to 90%”. At a 90% target retention, the scheduled interval in days comes out equal to the stability, which is why the two numbers look interchangeable until you change the target.

The power shape matters because it has a long tail. Plug in a card with 10 days of stability, due at day 10:

Missing a review by a week is not the catastrophe the classic exponential-decay diagram implies. This is one reason Moji has no daily review cap: letting a backlog clear late is far better than never clearing it, and the math says late reviews are still mostly recoverable. More on that in why we removed the review cap.

Why memory bands are defined relative to target retention

Once you have a recall percentage, you need thresholds to say anything useful about it. The tempting move is to hard-code them: solid above 90%, weak below 70%. Moji does not do that. The bands are defined as offsets from your target retention:

• Solid: retrievability at or above the target. The card is inside its schedule.
• Holding: within 15 percentage points below the target. The card is due, and this is exactly the state FSRS is aiming for at review time.
• Fading: more than 15 points below the target. The card is overdue far enough to be at real risk.

At the default 90% target that lands the boundaries at 90% and 75%. But the definition is the offset, not the number. If the target moves, every band moves with it, and “solid” keeps meaning “inside its schedule” instead of quietly meaning “better than some number we picked in 2026”. Fixed thresholds would silently mislabel cards the moment the target changed: at an 80% target, a card sitting at 82% is comfortably on schedule, yet a hard-coded 90% cutoff would flag it as weak.

The same principle keeps every surface honest. One threshold definition feeds the recall percentage on word rows, the recall sorts in word lists, and the order of the due queue, so a card cannot read as healthy in one place and at risk in another.

Why bands beat raw intervals

“Next review: in 12 days” tells you what the scheduler decided. It does not tell you how well you know the word, and it is not comparable across cards. A 12-day interval on a young card and a 12-day interval on a mature card mean completely different things about your memory.

Retrievability is comparable. Every card in your collection is on the same 0 to 100% scale regardless of age, difficulty, or how many times you have seen it. That makes a few things possible that intervals cannot support:

• Sorting a word list by lowest or highest recall to see what is actually slipping.
• Reading a single word’s history as a curve rather than a list of dates.
• Ordering the due queue by predicted recall, so an abandoned session still rescues the cards that were closest to being saved.

Moji shows this on each word’s detail screen as a recall chart: the recorded past drawn from your actual review log, sawtoothing back to full strength at every successful review, then a dashed projection forward to the scheduled date, with grid lines at 80%, 90% and 100%. For 食べる (たべる, to eat), you see the exact shape of your own memory for that word, not a generic illustration.

Why Moji waits two reviews before showing a number

A card you have swiped once has essentially no evidence behind it. FSRS will happily produce a retrievability for it, and that number will read as a near-perfect 100% while telling you nothing. Moji withholds the recall percentage until a card has at least two reviews, so the number only appears once it means something. Those very young cards still lead the due queue, on the reasoning that the least established cards are the ones a short session should protect.

When should you review Japanese vocab, in practice?

Open the app and clear what it hands you. That is genuinely the whole answer, and the reason is everything above: the due date is not a suggestion or a habit-building nudge, it is the solution to an equation that says “today is when this word hits 90%”. Reviewing early wastes the effort that makes memories stick, since a card recalled at 98% teaches your brain almost nothing. Reviewing very late is survivable but slower, because each lapse resets stability hard.

If you want the mechanics of how the schedule itself is computed, including how Moji’s two-button workflow (swipe left for again, swipe right for good) maps onto FSRS grades, see how Moji’s FSRS scheduler works. If you are comparing systems, the Anki alternative page covers where FSRS in Moji and FSRS in Anki genuinely differ, and how long it takes to learn Japanese puts these per-card numbers into a calendar.

Moji is an iPhone-only app with over 7,400 hand-curated Japanese words, every one with a curated example sentence, and it shows you the memory math rather than hiding it. Moji is on the App Store.

Sources

• FSRS algorithm specification, Open Spaced Repetition: https://github.com/open-spaced-repetition/fsrs4anki/wiki/The-Algorithm
• Retrievability values in the table computed directly from the FSRS-6 forgetting curve as implemented in Moji, with w20 at the FSRS-6 default of 0.1542.