I have been using Anki on a deck for two years, targeting a 95% retention with FSRS, but over the last year my retention has hovered around 93% (see screenshot and additional context below). Does anyone have any suggestions on what could be going wrong or how to fix this?
all the decks are set at 95% retention, and I am positive that I am rescheduling re-optimizing the parameters.
Maybe I’m just unlucky, and I’m one of the few people who still doesn’t adapt well to FSRS after the last update. Or, maybe FSRS doesn’t adapt well to large decks that are 1+ years old.
When I see posts like yours – asking why you aren’t reaching your Desired Retention (DR) – the simple answer I want to give is: “You’re getting too many answers wrong.” While that might be something you can fix with FSRS, it also might be something about your study practices, or just that you’re reaching for a retention level that is generally higher than anyone needs to get to.
When you get an answer wrong, what do you do to “relearn” the card and give yourself a better chance of getting it right the next time? If you just click Again and hope for the best, you’re missing a step.
Did that help? Generally FSRS schedules cards better when it has more review history to optimize over. There will always be easier cards and harder cards, so it helps if FSRS can see them.
Did you consider fixing or suspending your low-quality cards instead?
When I see posts like yours – asking why you aren’t reaching your Desired Retention (DR) – the simple answer I want to give is: “You’re getting too many answers wrong.” While that might be something you can fix with FSRS, it also might be something about your study practices, or just that you’re reaching for a retention level that is generally higher than anyone needs to get to.
Maybe. SuperMemo recommends 95% retention, and the statistics I have seen seem to support this number. Targeting retention lower than 95% seems to lead to forgetting too much, which defeats the purpose of using spaced repetition long-term.
When you get an answer wrong, what do you do to “relearn” the card and give yourself a better chance of getting it right the next time? If you just click Again and hope for the best, you’re missing a step.
I just relearn and leech at 4.
In the past, I would create new context around every card that leeched to support learning. But, that did not seem to do much, so I have gotten lazy and stopped that. Now when a card leeches and suspends, I just reset it. I only make adjustments to the card or create mnemonic when a card gets suspended twice or the card relates to numbers.
Did that help? Generally FSRS schedules cards better when it has more review history to optimize over. There will always be easier cards and harder cards, so it helps if FSRS can see them.
Yes, this got me from 90-91% to 92-93%.
Did you consider fixing or suspending your low-quality cards instead?
I leech and suspend at 4, but I reset most of those cards right after. I don’t really see cards get suspended more than twice, so I don’t think its necessarily a card quality issue (if this assumption is incorrect, I would be eager to know). However, I will look more into this to be sure.
I’m not familiar with that recommendation – from them or in general for spaced repetition studying. Have you considered that they use that term differently than we do in Anki? Here’s a good explanation of how Desired Retention and retention contrast with Retrievability (R). If you’re targeting 95% overall Retrievability – you don’t need your DR to be nearly so high.
Does that mean “I just click Again”?
You Reset it to New?! That’s unusual and probably ill-advised. You’re robbing FSRS off all that juicy review history to learn what your memory patterns are.
It seems like you’re probably teaching FSRS that over the long-term study of any given card, you’re always successful – since you make the unsuccessful cards disappear from view. Giving FSRS a false impression of success would make it schedule your intervals “too” long, and you’d have lower retention results – exactly the problem you’re trying to address.
Thanks. I was targeting a desired retention amount of 95%, not retrievability. Reading the Supermemo guide, you are correct, they use retention to mean retention, which I was not aware of. I also did some math that supported 95% as an ideal desired retention target, so I will have to double-check those calculations to see if they still hold up.
Yes
Interesting. I wonder if this is the main driver of me not reaching 95%. I will stop resetting cards.
My point was that scheduling to have a 95% chance of getting a card correct when it is due means you have a higher chance of getting it right the rest of the time. That’s usually absurdly more work than is necessary, but you do you.
Well, as I said above – you’re missing a step. If you don’t put in the work to learn/relearn the material, Anki can’t do much to help you memorize it.
