Many more reviews than expected with new card settings

[aoanla]

I tried searching but couldn’t find good terms to get results, although I am sure people must have had this problem.

The rule of thumb for reviews and new cards seems to be “you review about 10x your new cards”.
(And I assume the number of reviews is supposed to asymptotically approach this over time, with some noise.)

After 11 days of using Anki, with a new card limit of 20, my current studied cards for today is 352 - it’s taken me almost an hour to get this far, and there’s still 5 cards that I’m relearning that I’ve not finished relearning.
This isn’t a blip - the number of studied cards per day has been increasing day on day.

Is this just a teething problem, and my studied cards will actually go to about the level expected (~200, and about 30 mins max for study time), or is there something wrong here?
If there is something wrong… what should I do about it?

[Danika_Dakika]

It could just be growing pains, but there are things a user can do to veer them away from that general guideline –

1. Which scheduling algorithm are you using – SM-2 or FSRS?

2. What are your settings for that algorithm? Did you stick close to the defaults, or make a lot of changes?

3. What are your learning/relearning steps?

4. Do you graduate all of your Learn/Relearn cards to Review each day?

[aoanla]

I’m using FSRS as of 2 days ago (when I was definitely over 1000 reviews total).
I just did the “click optimize/check the estimate looks good” approach for FSRS, I assume it can do the right thing with the fitter.
(I’ve not manually adjusted anything else)

My relearning steps are apparently 10 min.

FSRS seems very wrong about what it thinks I should be spending as time though - the simulator says I should be spending about 33min/day approximately reliably from today for every day, and I am at twice that)

My max reviews a day is 200, I should say - but new cards take a lot of views for me to fix in my memory, and many cards I “forget” need a few repetitions too.

My true retention for the past “month” (11 days of course at this point, but it’s the longest sample I have) is only 64% - my desired retention in FSRS is 90%.

I do do all of my learn/relearn cards all the way to “learned” if that’s what you mean by graduating them to Review each day.

[aoanla]

Also, replying to myself: I see a rare few anecdotal posts dotted around that suggest that with less than 10 new cards a day, some people see up to 200 reviews a day (that is more than twice the “rule of thumb” that’s widely distributed that it’s x8 to x10 your new cards).

How reliable is this rule of thumb?

[sorata]

@aoanla

How reliable is this rule of thumb?

Not so much:

@dae

I seem to recall coming across users in the past with a review ratio close to 20x new cards.

@aoanla

FSRS seems very wrong about what it thinks I should be spending as time though - the simulator says I should be spending about 33min/day approximately

As you already said, your retention is very low and you’re doing a lot of intraday reviews, so it’s expected that you’re taking much longer than FSRS estimates. In the latest beta versions, Anki will also take into account the (re)learning steps so perhaps that’ll get somewhat more accurate.

@aoanla

My true retention for the past “month” (11 days of course at this point, but it’s the longest sample I have) is only 64% - my desired retention in FSRS is 90%.

One thing is, the higher the DR you set the more reviews you get. As your true retention was quite low, you’ll probably get more reviews now with FSRS. I suggest you try lowering the number (and see how much decrease in daily load you get using the simulator).

[aoanla]

Okay, so I’m just going to ignore what FSRS says (it, hilariously, on evaluating the settings I have, thinks I should push my Desired Retention up to 93%) and instead turn down DR?

[sorata]

Are you talking about compute minimum recommended retention (CMRR)?

[aoanla]

yeah, that one

[sorata]

I am not sure what to recommend then. CMRR is being reworked and is the topic of some debates currently. I guess just keep the DR at that level (90%) and experiment lowering other things (new cards/day) in the simulator.

[aoanla]

Sure, but since the simulator is already wrong (it thinks that right now I should be spending ~30mins a day, and I’m spending twice that), how do I work out what to expect versus what it incorrectly thinks?

[sorata]

Look at number of reviews, not time. Also, update your Anki if you haven’t already.

If you’re fine with beta version (probably only minor bugs):

Release 25.06b1 · ankitects/anki

[aoanla]

Hm, interestingly with 25.06b..

Firstly - the optimiser finds an entirely different set of parameters (so, yeah, those changes did something).

Secondly - the simulations still imply that with a Desired Retention of 90%, I would need to go down to 4 New Cards per day to keep my average revision time per day at around my target 30 minutes/day. They also continue to agree with the old implementation that I need to increase my Desired Retention to get my average revision time per day to be lower - changing it from 90% to 95% (and keeping new cards at 20) brings down the average revision time from the 40-50min a day to about 30 min a day… but significantly reduces the number of cards it thinks I will memorise over time.

(Indeed, the simulator thinks that, even with a desired retention of 70%, it will take me until 2032 to memorise just the starting 5000 cards in the deck I have… and with the settings necessary to have 30min of revision a day, I won’t memorise more than 4000 by the end of 2034 - and will take > a decade to memorise the entire deck. )

Does this seem reasonable? Is there something wrong with my brain?

[sorata]

@aoanla

Firstly - the optimiser finds an entirely different set of parameters (so, yeah, those changes did something).

It’s a new version of FSRS, FSRS-6. The number of parameters have also changed.

@aoanla

I need to increase my Desired Retention to get my average revision time per day to be lower

If you’re talking about CMRR, that’s the wrong interpretation. CMRR tries to lower other metrics, not review load.

@aoanla

changing it from 90% to 95% (and keeping new cards at 20) brings down the average revision time from the 40-50min a day to about 30 min a day

It’s possible given that at lower retention rates, you forget more cards and those have to go through relearning. Maybe you spend lot of time relearning.

@aoanla

but significantly reduces the number of cards it thinks I will memorise over time.

@L.M.Sherlock Do you know what’s happening? Increasing DR should increase memorised count, right?

@aoanla

to memorise just the starting 5000 cards in the deck I have

FYI, that’s more complicated than that. FSRS takes the probability of recall of every card and use their sum to estimate how much you’ll remember. And that doesn’t mean you’ve learned the first 5K.

@aoanla

I won’t memorise more than 4000 by the end of 2034 - and will take > a decade to memorise the entire deck.

To shed more light on this, FSRS wouldn’t be showing you cards until the probability is low enough (with DR 70%, you forget 30% when you review). So, over time you see cards less and less frequently just to maintain this forgetting rate.

And also at really long intervals, FSRS probably would sort of break down. I mean no one really learns things for multiple decades. At some point, and as long as it’s not obscure trivia the cards will just become unforgettable.

[aoanla]

@sorata

@aoanla

I won’t memorise more than 4000 by the end of 2034 - and will take > a decade to memorise the entire deck.

To shed more light on this, FSRS wouldn’t be showing you cards until the probability is low enough (with DR 70%, you forget 30% when you review). So, over time you see cards less and less frequently just to maintain this forgetting rate.

No, to be clear, “the settings necessary to review for only 30 minutes a day” are with DR at 95%. Increasing DR significantly both: reduces predicted review time and reduces the rate at which it predicts I will memorise cards. (If I hold new cards / day constant.)

[sorata]

Maybe a review limit cap isn’t allowing all the cards to be reviewed (in the sim). So, some cards are stuck in a backlog which causes this low memorised count. Not sure about the time one though.

@Expertium If you got anything to add (already pinged Jarrett).

[aoanla]

So, I upped the maximum number of reviews per day in the simulator - and yes, that’s one of the things that’s causing weird results… but I really don’t like what it thinks will happen if I effectively uncap the max reviews by setting it to a v large number like 1000.

Then, for any Desired Retention - at least between 70% and 95% - if I add 20 new cards per day it thinks I will rapidly grow to needing more than 700 reviews per day by November (monotonically increasing), and will need to spend more than 2 hours on Anki daily by October.

I have to pull new cards per day all the way down to 3 to get it to predict me an uncapped reviews simulation that gives me 30 minutes of Anki time predicted per day (at about 200 reviews per day).

[Danika_Dakika]

I’m not going to answer any questions about the simulator, because (1) it’s experimental, (2) it’s only a simulation, not an absolute certainty, and (3) you have very little review history for it to rely on (and what you do have sounds like it’s been a bit turbulent). Rather than try and troubleshoot the simulator, let’s look at your actual settings and see how to get to closer to what you expect.

@aoanla

My true retention for the past “month” (11 days of course at this point, but it’s the longest sample I have) is only 64% - my desired retention in FSRS is 90%.

This is the biggest indicator of the issue. Your workload is high because you’re only getting 64% of your answers right, and you asked FSRS to schedule your cards so you’ll get 90% right. That’s a big gap! Anki will keep your cards scheduled on shorter intervals (which means higher workload) while its pushing you toward the goal you set.

@aoanla

but new cards take a lot of views for me to fix in my memory, and many cards I “forget” need a few repetitions too.

Because your collection is still so new, and True Retention isn’t counting your outcomes on short interval Learn/Relearn cards – I think this is the other piece of the puzzle.

As language learners, it’s not unusual for us to use premade decks where we don’t necessarily know any of the words. But if we don’t do the work to learn those words before introducing them in Anki, then we need to do that work when they are introduced and we miss them the first time. If all you do with a New card is check the answer and grade it Again – you’re skipping that step.

When you get a new card wrong, you need to stop and learn it. For me, that means checking my dictionaries to make sure I understand the word, finding example sentences so I understand how it would be used, thinking of mnemonics to differentiate it from other similar words, etc. Steps like that will help improve the chance I’ll get it right the next time – and that’s a MUCH better use of my time than studying the card 10 more times today, trying to brute-force it into my memory. And it’s the same with lapsed Review cards – figure out why you got the card wrong, and solve it so you can get it right the next time.

@aoanla

with a new card limit of 20, my current studied cards for today is 352

I think you should turn your New cards down a lot (or to 0) until you get a handle on this. But it matters where that 352 is coming from. Is that your number of “reps” for the day (the number of times you did a front-back-grade-the-card “study” action), or the number of unique cards you studied for the day?

If you’re looking in Stats at Today or Calendar or Reviews, they are giving you reps. If you click on that square of your calendar, it will open the Browse window, and you can see how many unique cards you actually studied.

A couple more things that aren't necessarily your main issue, but bear mentioning ...

@aoanla

My max reviews a day is 200

If you are capping your daily reviews at less than your actual number of due cards, you are postponing the rest of the cards. It’s essentially a hidden backlog of overdue cards. When you study them overdue, it means your memory has decayed further, and you’re less likely to get them right, which results in lower retention, and more time spent rehabilitating those cards.

It sounds like you changed it in the simulator, but if you are already hitting that daily limit in your actual decks, you are better off changing it to unlimited (9999), and addressing workload issues more directly.

@aoanla

How reliable is this rule of thumb?

For a beginning user who is sticking close to the default settings, it’s generally quite reliable. Just like Damien talked about in the other thread, I’ve seen outliers now and then. But after only a couple weeks of Anki, I think it’s too soon to assume you’ll be one of them.

[aoanla]

@Danika_Dakika

I think you should turn your New cards down a lot (or to 0) until you get a handle on this. But it matters where that 352 is coming from. Is that your number of “reps” for the day (the number of times you did a front-back-grade-the-card “study” action), or the number of unique cards you studied for the day?

That’s the number of reps (the value reported as “cards studied” by Anki). The number of unique cards is smaller (and less than 200, the review limit).

Part of the problem here, from reading your whole post, is that apparently I simply can’t expect to spend 30 minutes a day with Anki where some cards are new to me and get anywhere near my goals for retention. This is demoralising, but that’s not Anki’s fault.

I will say that it’s not entirely true that the cards are new to me - I’ve been studying the target language I’m learning for > 1 year already, and I’ve usually met quite a few of the cards in this deck before. Some of them are very easy for me. Some of them - even ones I’ve met before - are exceptionally hard, it’s not just the ones that are new to me entirely that I have trouble with retention on. The deck I got actually has example sentences for each word (when you fail it) and audio, but even reading those each time/relistening, I have an abysmal retention rate for the cards that aren’t “easy”.

[Danika_Dakika]

@aoanla

That’s the number of reps (the value reported as “cards studied” by Anki). The number of unique cards is smaller (and less than 200, the review limit).

That’s good. You can see how if you don’t have to study the cards over-and-over so many times, your workload drops considerably.

@aoanla

Part of the problem here, from reading your whole post, is that apparently I simply can’t expect to spend 30 minutes a day with Anki where some cards are new to me and get anywhere near my goals for retention.

It wasn’t my intention to say that! Of course you can stick to 30 min per day, and you can introduce new cards – but for you, it sounds like 20 isn’t the right New card limit. You just aimed too high starting out.

It will take some effort to get back to a better baseline, and then you can start adding New cards again – just at a slower pace. You might want to divide your current Review cards up, and put some of them on hold for a little while, so you can work on getting your collection stable bit-by-bit.

@aoanla

I have an abysmal retention rate for the cards that aren’t “easy”.

Don’t get stuck in thinking that you can’t learn those cards. You just haven’t learned them yet. But it sounds like you’ve learned them in the past, so you need to remember that it’s not impossible.

[aoanla]

@Danika_Dakika

@aoanla

Part of the problem here, from reading your whole post, is that apparently I simply can’t expect to spend 30 minutes a day with Anki where some cards are new to me and get anywhere near my goals for retention.

It wasn’t my intention to say that! Of course you can stick to 30 min per day, and you can introduce new cards – but for you, it sounds like 20 isn’t the right New card limit. You just aimed too high starting out.

It will take some effort to get back to a better baseline, and then you can start adding New cards again – just at a slower pace. You might want to divide your current Review cards up, and put some of them on hold for a little while, so you can work on getting your collection stable bit-by-bit.

I didn’t quite mean that you said that explicitly. However, the inference from the discussion is that:

1. if I see a new card, I need to spend a lot of time (probably minutes per card) to internalise it
2. I also need to do at least as much language learning stuff as I was doing before in addition to Anki

So

1. by itself implies that at 20 cards/day, I need 20minutes just for the new cards to memorise them etc
2. means that I also need a lot more additional time - not less time! - than I was using before, so my “total language learning” time per day goes up relative to without Anki.

I think a natural consequence of this is that if I want to use Anki, and scale new cards accordingly, I can’t expect to do much more than 5 new cards a day (considering that 200 cards in review is probably already about 30 minutes without considering news, and only if I review each card once).

5 new cards a day at 5000 cards total is 1000 days to have seen the whole deck, which is 2 and 3/4 years.

I would like to have a functional vocabulary in less than 2 and 3/4 years of my life

That’s what’s demoralising - that apparently I am just not going to be able to do useful stuff in my target language for ever. (That’s going to be about 4 years after I started trying to learn this language that I can actually do anything useful in it, including the pre-Anki period.)