hello,
after asking many IA bots, after running the FSRS simulator (which I think is much too optimistic), I still have the same question : if I stop adding new cards, will my daily number of reviews keep decreasing indefinitely ?
I have 40000 cards, a desired retention rate of 86%, a real retention of 88%. one of my decks is 3 year old and I didn’t see much of a decrease last year.
what is older users experience on this ?
thank you !
Yes, it will generally decrease, as long as you’re reaching your Desired Retention (DR). But even if “indefinitely” could be mathematically accurate, it’s a bit of an exaggeration for any practical purpose. As you have fewer and fewer cards due, the decreases will be slighter and slighter.
What is the simulator telling you that looks wrong? It’s an experimental feature, so it might not fit everyone perfectly.
Are you using the right figures in your simulations?
Linking to your prior threads about the FSRS Simulator –
let me give you two examples :
deck 1 : 12000 cards, currently 37 reviews a day, lots of mature cards. only a very light decrease in the past year and the simulator tells me there will be a massive drop in reviews to reach 23 in 2 months and 17 in a year. I can tell you from experience that the big 2 month drop will not happen and 17 is also super optimistic.
deck 2 : 14000 cards, currently 55 reviews a day, many more young cards and instead of a big drop (thanks to these young cards), I see an increase in the next 2 months that is coming out of nowhere and then another bump (much easier to understand because I added a bunch of new cards recently and they will resurface in about 3 months) and then a downslope that seems fairly plausible.
I’m too lazy to keep my data and see if the predictions will be correct but considering what happened in the past, I bet they’re very wrong.
what is the feedback from other users ? they think the simulator is pretty correct ?
in the simulations, I’m not changing any number and adding new cards/day is not even working in deck but it is in deck 2. very strange
I’m trying to look at your collection that is synced ot AnkiWeb, but I’m not finding anything that quite matches what you’re describing. Is your AnkiWeb collection up-to-date? What decks/presets are you using in those examples?
deck 1 is ENGLISH, deck 2 is РУССКИЙ.
I think everything is synced and up-to-date. I didn’t even know you had access tot my decks, that’s funny 
There’s just a few of us who do – for support purposes. And we only use when there aren’t any better options.
Today, I’m mostly using your data as an exercise to see if there really are significant problems with the FSRS Simulator. You shouldn’t expect this kind of detailed review regularly.
This deck/preset has less than 9500 cards, 99%+ are Mature. New cards have trickled in very slowly, but almost all introduced now. Average workload for the next month is 26 cards/day – avg. reps for the past month is 33/day.
DR is 86%, but retention is higher –
So, this projection doesn’t look unrealistic to me. That “massive drop” might just be the natural progression as you study more of those long-interval cards and get them shifted to even longer intervals. And the slight bump after that could just be the result of the spike you had in Learn reps ~2 months ago – perhaps you were introducing more New cards for a week or so?
This deck/preset only has ~8600 active cards, and about 98% of those are Mature. [There are also ~5K New, but 0 introduced for several months.] Average workload for the next month is 43 cards/day – avg. reps for the past month is 62/day.
DR is 86%, but retention is quite a bit higher –
Here, your median interval of 9.83mo is longer than your median S of 5.92m, and your Avg. R is 92% – much closer to what you’d expect with a DR of 86%.
But if you look at your Review history, you had a huge spike of daily workload 2 months ago, for about a week (perhaps backlog catch-up?).
A burst of studying like that often has a cyclic effect on your workload – which can echo for the next 2 or 3 study cycles as it dissipates/load-balances away. Given that, I don’t think this projection is all that unreasonable either.
that’s perfectly fine and you’re being very helpful
what does median stability consist of ? could you please explain why it’s unusual ?
the DR used to be much higher before. a year ago, I’m not sure but I guess it was even higher than 90%. I decided to lower it because I wanted to lighten the workload and chatGPT kept telling me to do it and that I was reviewing too much with too many good guesses (more than 95% on most days).
I see that I added about 5 new cards (not a lot in my opinion, I keep adding 1 new card every day on average, sometimes just a correction of previous cards that were slightly wrong/not specific enough) every day for a week 60 days ago
my mistake, I forgot to subtract the new cards, not active yet.
exactly !
You’ll find it at the bottom of your Stats > Card Stability graph.
Median: Median - Wikipedia
Stability: ABC of FSRS · open-spaced-repetition/awesome-fsrs Wiki · GitHub
FSRS calculates a card’s Stability (based on 90% Retrievability), and then increases or decreases from there based on your DR to set the card’s interval and schedule the next review. Since your DR is turned down to 86%, you’d expect your interval for each card to be quite a bit longer than its Stability. The median just gives you a convenient way to compare the 2 sets of data.
That’s still 5x the usual number, even if the usual number is a small one.
thank you very much !
do you have any idea why, if I change the number of new cards/day, the simulator seems to work for my Russian deck but not for the others ?
for my Mandarin deck, when I add X number of new card/day, the graph doesn’t move an inch. it’s always exactly the same. the number of new cards doesn’t affect the curve.
for English and Spanish, the graph is roughly the same but with daily variations, so the bug is different.
for my deck called KUKU LE SHBOUNK (it was very few cards), it gets totally nonsensical.
1 = 0 new cards, 2 = 1 new card, 3 = 2 new cards : it looks pretty normal to me, the curve keeps going up regularly.
4 = 3 new cards : it starts to become weird as the curve goes back down after an initial rise.
5 = 10 new cards : same as above but more extreme. it even says that after a year, I would have fewer cards to review than with lines 3 and 4.
(same number of new cards for each curve on every graph)
The Mandarin deck’s preset/limits are probably blocking new cards, or maybe you don’t have new cards that aren’t suspended.
The English and Spanish decks probably have too many cards (12000?) for a small change of 1-10 new cards per day to make a big difference. Still feels bizarre.
For the small deck, what’s probably happening is that with higher new cards per day, you finish the deck sooner, but have a steeper start. The peak is when no new cards are left.
ohhh, do you mean that the number of new cards is limited by the actual number I have in my decks ? that would explain everything. English and Spanish have 4 new cards each, Mandarin doesn’t have any. in Russian, the number of new cards is about a third of the total number of cards. in Kuku le Shbounk, the number of new cards is four fifths of the total number.
I guess the reason is those decks have almost no new cards. I was thinking that the graph was adding X cards everyday, no matter how many you have in your actual decks (because we’re all adding new cards regularly).
Yes. There’s the “Additional new cards to simulate” option for a reason.
ohhh, now I feel stupid. got it, thank you !
