Changing FSRS Settings Resets Learning Cards' Progress

Hi all, was curious if it’s normal for a card’s learning progress to reset after adjusting the FSRS parameters. I do see this in the manual (“Deck Options → Reschedule Cards on Change”): “…future reviews will use the new scheduling…”, so maybe?

For context, I do use learning steps greater than 1d (despite it not being recommended. Due to my bird brain, I know I won’t remember cards if I only see a learning card once before the graduating interval is set to something like 14 or 21 days).

I’ve been using the following learning steps with FSRS: 2h 1d 2d 3d. Just yesterday, I decided to add 11d as a final learning step. When doing my reviews today, I noticed that current learning cards had their progress “reset”, i.e., their next interval is 2h regardless of which learning step they were previously on. Example below:

Is this intended behavior? How can I prevent this in the future and/or fix the intervals/steps of my current learning cards?

As I mentioned above, I don’t feel comfortable with only having learning steps <1 day, as that’s not sufficient repetition for me to retain info. My desired retention is already set to 95%. I also can’t really afford to take the time to try using only <1 day learning steps, study a bunch of cards, and let the algorithm optimize its parameters over time.

Any thoughts or help would be appreciated. Thanks!

I don’t think this is related to any changes you made in FSRS. You changed the learning steps. Depending on what version you’re using, that can cause cards in Learn to skip or repeat steps. See: Repeated/skipped learning steps .

Thanks for your reply. I took a look at the thread you linked, and based on Dae’s reply (marked as solution) it seems like the “resetting progress” shouldn’t be happening, though.

I even used the “Grade Now” to simulate answering the card and can confirm that learning cards are indeed being placed back at the first learning step:

This card was last answered on 1/30 before I adjusted the learning steps, so when I answered it today (2/4) I was expecting the next interval to be the newly added step of 11 days, but as you see in the screenshot, it reverted to 2 hours, and then with the sequential “Grade Now”, it progresses through all learning steps.

Did you check what version of Anki you’re using though?

25.07.5

That fix should have been included in 25.07. When did you update to that version?

Don’t remember exactly when, but definitely long before I encountered this issue and made this forum post.

I don’t know if cards already in Learn when you got the fix would suddenly behave correctly. It seems more likely they would start the learning steps over – which is what you’re seeing. Since you set your cards up to stay in Learn for an exceedingly long period of time, that seems like it could be the issue.

Sorry if I’m just misunderstanding, but I thought the point of the fix was to prevent cards from returning to the first learning step if you adjusted the number of learning steps.

The fix makes sure a card entering Learn has appropriate landmarks so that it won’t skip or repeat steps when the steps change. But if you have tons of cards already in Learn, you probably need to help them out.

For cards that have already had plenty of reps in Learn – you can use Set Due Date to graduate them to Review and let FSRS take over their scheduling. But keep in mind that you’ve overloaded these cards with correct-answer review history while holding them down at your own short intervals. FSRS will consider all of those (one of the reasons why using long steps is not recommended with FSRS).

For cards that are recently introduced (just a couple reviews), you can reset those to New.

Got it, thanks. So just to be clear, if a card is already in Learning, adjusting the learning steps will “reset” their learning steps progress, and this is an intended feature. Is that right?

It won’t necessarily “reset their learning steps progress” – but it is likely to cause skipped or repeated steps. No, that’s not intended, it’s bug that has been fixed.