Desired retention in FSRS: what 90% means and when to change it

One number in your deck settings decides how often almost every card comes back: desired retention. Turn it up and you'll forget less, at the cost of a heavier review queue. Turn it down and the queue gets lighter, at the cost of forgetting more. Here's what the number means and how to pick it.

What "90% retention" actually means

FSRS — the scheduler behind R3CK0N's flashcards — models memory as a forgetting curve: for each card, the chance you'd recall it right now drops smoothly over time from the moment you last reviewed it. Desired retention is the recall probability you want a card to still have on the day it comes due. At the default 90%, FSRS times each interval so that, on average, you have a 9-in-10 chance of getting the card right when it reappears.

It's a target, not a guarantee. Some cards you'll get right well past that point; some you'll forget early. Ninety percent is the average FSRS is aiming for across your whole deck, and it recalculates the exact interval for every single card from that card's own stability — a well-learned card gets a much longer gap than a shaky one, both aimed at the same 90% odds.

The trade-off, in one direction

Raise desired retention and FSRS schedules reviews sooner, because it takes less elapsed time for recall odds to fall to a higher floor. Lower it and reviews are scheduled further apart, because there's more room for recall to drop before it's due. Neither direction is free:

Roughly how much workload changes

The relationship isn't a straight line — going from 90% to 97% costs a lot more than going from 90% to 95% does, because intervals get compressed fastest at the high end of the curve. As a rough, order-of-magnitude feel for it relative to the 90% default (your own numbers will vary with your parameters and material):

R3CK0N keeps the setting between 70% and 97% for this reason — outside that range the trade either barely reduces workload or demands an unreasonable number of reviews for a small gain in recall.

When to move off the default

Push it up toward 95%, close to a deadline

If an exam is two weeks away and you'd rather over-review than risk forgetting, a higher retention target front-loads the extra reviews into the time you have. It's a short-term dial, not something to leave on permanently — the review load it demands doesn't ease off on its own.

Pull it down toward 80%, for a deck you're maintaining for years

Vocabulary you're keeping "just in case," or a huge deck you never intend to finish importing all at once, benefits more from a light, sustainable trickle of reviews than from squeezing out the last few percent of recall. Forgetting an occasional card and relearning it costs little when there's no deadline.

Leave it at 90% for almost everything else

It's the default for a reason: a reasonable middle ground between how often you'll forget something and how much time reviewing costs you. Most decks never need to move it.

Changing it doesn't erase progress

FSRS keeps tracking each card's own memory state — its stability and difficulty — no matter what the retention target is set to. Changing desired retention only changes the target FSRS aims the next interval at; it doesn't reset or discard anything about cards you've already studied. Lower it and a card's next interval gets longer from wherever it already stood; raise it and the next interval shortens the same way.

Retention and new-cards-per-day pull in the same direction

Desired retention isn't the only dial that sets your review load — new cards per day does too, and the two multiply together rather than sitting side by side. Raising retention on a deck that's already getting 30 new cards a day compounds a heavy schedule into a heavier one; the same retention bump on a deck getting 5 new cards a day is barely noticeable. If your review queue is already uncomfortable, check which of the two settings is actually driving it before changing either — lowering new cards per day is usually the gentler fix, since it only slows how fast the deck grows rather than reshaping every existing card's schedule.

A quick way to feel the difference

If you want to see the effect rather than take the table's word for it, try a small deck at two extremes for a couple of weeks each: run one at 80% and watch how much sooner intervals stretch past a month, then run a similar deck at 95% and watch how much more often a card resurfaces inside a week. The gap between the two is the whole trade-off in miniature, and it makes the number on the settings screen feel a lot less abstract.

Frequently asked questions

Where do I change desired retention?

In the deck's settings, from the deck editor. It's per-deck, so you can run one deck hot before an exam without touching the rest.

Does a lower retention mean the algorithm is "worse"?

No — it's the same model aimed at a different target. Lower retention decks forget more often on purpose, in exchange for a lighter review load.

Should new and old decks use the same retention?

Not necessarily. It's fine to run different values across decks depending on how much each one matters right now.