Review intervals: too many cards with too long intervals

Anki FSRS
1

Hi, Anki learners, I was checking my stats and I found something quite odd in one of my decks. I have unproportionately many cards with very long repetition intervals, see the picture below. The graph seems more intuitive for other decks, the number of cards with longer intervals kind of fades out. My question is, do you think this is ok and if not, is there some way how to fix this? Thanks

P.S.: I switched to FSRS some time ago and I probably wasn’t using the again button when I should have in the past.

image
image756×467 19 KB

2

That sort of wall is almost always caused by a max interval. 5 years * 365 = 1825 days – is that your max interval for the preset this deck uses?

3

Thanks for your answer @Danika_Dakika . Yes, the max interval is indeed set to 1825 days, but the distribution of the interval lengths seems off. Below I am showing the same graph for a different deck, which I think is more appropriate. The problem with the previous graph is that when I see the cards with the max interval, I don’t really consider them that well remembered. I suspect something happend when I switched to FSRS. Do you have any suggestion how to fix this or should I just reschedule them manually?

image
image755×469 20.1 KB

4

Ah, I see – inside your max-interval question was a hidden FSRS-parameters question. :sweat_smile:

Are these decks assigned to the same Options preset, or a different one?

If you have just one preset where your parameters seem off, we’ll also need to know –

  • Your parameters as text, and your Desired Retention (DR).
  • How long has it been since you re-optimized your parameters?
  • When you optimize in this preset, how many reviews does FSRS count? (If you don’t remember, you can click “Optimize Current” to get the number, and close without saving.)
  • Your Stats – click Save PDF to get the whole page and attach that here.
  • The Card Info for a card that you think is mis-scheduled for too long an interval (screenshot is fine).
5

exactly :slight_smile: and also I am afraid that my previous use of hard vs again buttons was not really exemplary

I have a specific preset for each deck, to your bullet points:

  • FSRS parameters for the problematic deck: 4.4362, 10.6435, 19.5524, 32.8239, 6.4330, 0.7300, 3.0314, 0.0011, 1.8535, 0.2845, 0.7799, 1.6582, 0.0268, 0.4785, 1.7977, 0.6161, 1.8891, 0.3510, 0.3037, 0.0100, 0.2306; my desired retention is set to 92% (for comparison, FSRS params for a correct deck with DS 90% are: 2.0400, 3.0491, 4.0316, 5.1145, 6.5415, 0.6433, 2.9684, 0.0918, 1.7260, 0.3619, 0.6548, 1.7004, 0.0327, 0.5513, 1.8298, 0.4487, 1.8872, 0.5249, 0.3269, 0.0100, 0.2031)

  • I optimize once in a few months, I believe it’s time to optimize again soon (FYI: after optimization I get new parameters: 5.3240, 12.4516, 22.4717, 37.1589, 6.4059, 0.8033, 3.0149, 0.0349, 1.8741, 0.1894, 0.8008, 1.6253, 0.0010, 0.4085, 1.7828, 0.5375, 1.8972, 0.4027, 0.2502, 0.0100, 0.2803)

  • when I optimize, the FSRS says it counted ⁨4,548⁩ reviews

  • pdf with my stats of the problematic deck: anki-stats-2026-08-28@08-47-16.pdf (311.5 KB)

  • example of info to problematic card (it could have been scheduled before FSRS was installed, but I am not sure, anyway the recent manual reschedule was done so that I could rate the card properly when it soon comes to review, I want to test how many such cards with max interval will be evaluated with again button):

    image
    image800×833 30.9 KB

Thanks for your help, I would love to understand better what’s happening and how to change my behaviour so that I will fix this in the future.

6

Yes, it definitely looks like there’s a problem. Your median Stability for this deck is 4.57 months, when your median interval is 4.78 years. At 92% DR, that shouldn’t be happening. But your Stability graph is counting less than 800 cards in your deck, even though you’ve got over 2K active. I can also see in that Card Info, the card doesn’t have a memory state.

This one had me stumped for a bit, so I took a look at your collection on AnkiWeb. The things that usually cure the missing-memory-state problem weren’t working (optimizing, Helper add-on rescheduling, etc.), but then I saw that you’re using an ignore-before date to exclude most of your review history. You were optimizing based on only 25% of your cards. That appears to be what was causing most of the inconsistencies in the data – "Ignore cards reviewed before" question - #2 by Danika_Dakika .

image
image777×118 4.15 KB

But getting all your history back does show that this is probably the root of your problem. Your parameters are cranked up for incredibly easy material, because you hardly ever grade cards Again. So – is this incredibly easy material? If it is, then there’s less we need to try to fix.

Before you try to cure your collection, I can give you a preview of what things might look like, because I’ve already done the first few things I would have recommended for you to start with –

  1. Run Tools > Check database.
  2. Remove ignore-before date and reoptimize.
  3. Select all cards in this deck and use the FSRS Helper add-on for Cards > FSRS: Update memory state and reschedule.

I’m sure there are things you’d like to change about this – but at least it gives you a baseline.

You can see that your Intervals and Stability start to make sense [even if they are still not what you want]. Your Retention and Answer Buttons show the issue better – you use Again 10-15% for on Mature cards, but you rarely use it for Young cards, and almost never for Learn/Relearn cards.

image
image1632×540 81.5 KB

image
image1578×468 72.8 KB

New parameters – optimized on 19K reviews: 3.5976, 13.3256, 33.1026, 71.8613, 6.2772, 0.7398, 3.5431, 0.0011, 1.9881, 0.1459, 0.9230, 1.6242, 0.0012, 0.5351, 1.7320, 0.2768, 2.0188, 0.6346, 0.5978, 0.0100, 0.4888.

That specific card is scheduled the same, but now it has a memory state. [Ignore the current interval – it’s from what you manually set.] A bunch of those cards with overly-long intervals have been rescheduled into the past, so you’ve got a lot of backlog to work through. But it’s really just moving those cards from sitting out there in the future as already forgotten time-bombs, to where you can actually deal with the problem now. [And we can talk about ways to manage a backlog efficiently.]

image
image732×842 18 KB
image
image780×603 22.5 KB

To start talking about adjustments –

  • What is going on with your grading history? How were you using Again, if not correctly? Were you using it differently in Learn/Relearn than for Review cards?
  • Is this deck really easy for you?
  • Can you handle having a backlog of ~850 cards, that you’ll need to chip away at?
  • Do you have other presets/decks with problematic review history? Or do you have other ones with similar material where the review history is more accurate?
7

Thanks for your thorough analysis, it’s very helpful! To your questions:

  • What is going on with your grading history? How were you using Again, if not correctly? Were you using it differently in Learn/Relearn than for Review cards?

    I think in the past on many occasions I used the button Hard in cases where in reality I should have used Again. And recently I was using Hard and Again buttons to setup a first interval for new cards (i.e., Learn/Relearn) to get the first review interval closer to something that looked more appropriate (after switching to FSRS, I was getting 3 weeks+ for the first review, which is not ok for the material in this deck, this is also the reason why I used the ignore cards option for the FSRS optimizer).

  • Is this deck really easy for you?

    Not really, it consists of technical stuff that involves a certain amount of thinking and non-trivial retrieval, so I consider it the toughest subdeck I have.

  • Can you handle having a backlog of ~850 cards, that you’ll need to chip away at?

    I believe it should be manageable, but if there’s a way how to add the backlog cards iteratively (over one month for example), I would definitely prefer this approach. It would be nice to be able to say for example: each day, give me 10 review cards from the backlog and the rest from the standard stash.

  • Do you have other presets/decks with problematic review history? Or do you have other ones with similar material where the review history is more accurate?

    This is the only problematic subdeck I have, there are two other that are up-to-date and that I use every day, these cause no trouble. There are few other old decks that I am trying to reduce and delete, but this is a process that will take some time, since I am trying to redo the cards contained in those decks. Make them more digestible.

I hope I answered all the questions satisfactorily and I am looking forward to the next steps that I could do. Thanks for your input, this helps me to scratch a long standing anki itch that I have :slightly_smiling_face:.

8

See what you think of these options. [This is for what you’d do after you’ve done steps 1-3 from my post above.]

Option A
Do nothing. Well, not really “nothing” – continue studying cards and grading them correctly, and re-optimize monthly. Work through the backlog created by the reschedule, and when you’re done with that, reassess.

If there are still a lot of cards scheduled for over-long intervals, you can pull those into a Filtered deck in batches and study them early. [Use a search filter like prop:ivl>X prop:due>Y – cards with an interval longer-than-X that aren’t already due to be studied soon (within the next Y days).] That will tell you if those are more time-bombs, or if they are just cards that deserve a long interval.

Option B

In that case, it might be appropriate for you to use the FSRS Helper’s “Remedy Hard Misuse” feature. It has it’s own pros and cons – see Reddit – but it seems like your grading habits were far enough off kilter that it would be worth doing. [For your situation, I wouldn’t recommend either of the other possible solutions in that Reddit post.]

Note that it has its own “Undo” feature (but make a fresh backup of your collection first, just in case) – and that you can limit it to just a certain time range, but not to just a certain part of your collection. If there are other decks that don’t have this issue (or that you don’t want to fix), there may still be a way to do this (using export/import and a temporary profile), but it will be more complicated.

The main benefit here is that it’s one-and-done. After that you’ll re-optimize and reschedule (using the add-on), and go about your studying.

Option C
Increase your DR – by quite a bit. As you continue studying cards and grading them correctly, you’ll be able to re-optimize and gradually lower your DR to normal, rescheduling (using the add-on) each time.

This is something there is no set formula for, so it will be up to you to monitor how your parameters and scheduling look, and adjust accordingly. The FSRS Visualizer is useful for that – and for comparing any before/after sets of your parameters on a level playing field.

9

After reading the reddit post, I have decided to go with the Option A, which seems to be the one, that addresses the problem of inappropriate review intervals in the most controlled way. All of them involve quite high amount of work, so from this point of view I wouldn’t be able to dodge that bullet :slight_smile: . Thank you very much for your thorough assistance with this. I will try to update this answer in one month to give you my final feedback.