FSRS - one of my decks is stuck in "ease hell"

Anki FSRS
1

Hello, I upgraded from Anki 2.1.22 to Anki 25.09.4 + FSRS about 4 months ago, and I’ve been loving it so far!

Most of my decks are fine, but one of my main decks is stuck in an “ease hell” type situation:

Most of the cards are basically stuck at the same interval even after several reviews. Many cards are stuck at an interval of about a month. Some cards are stuck at a really short interval, like 1 week.

Potential risk factors:

  • I’ve amassed a massive backlog in this deck a couple times. Currently I am in the middle of chipping through a backlog.
  • This deck is relatively old, dating back to when I first started Anki in 2019. I don’t think I’ve done anything particularly unusual with it, but who knows.

Luckily, I generally only use the good / again buttons and haven’t been misusing the hard button.

Here are my parameters:

1.0921, 4.1881, 10.6580, 23.6178, 6.7186, 0.5284, 3.4809, 0.0010, 1.5102, 0.5593, 0.4195, 1.6710, 0.0578, 0.5767, 1.9422, 0.6014, 1.8729, 0.9206, 0.6812, 0.0100, 0.5018

I hit the “opto all presets and save” button about once a month.

As for addons, I don’t think any of my addons should be relevant:

image
image899×496 26.3 KB

Here is my collection file (sans media). The “ease hell” deck is my main Japanese sentence mining deck called: 2. 日本語::a. Mined

Here are some representative examples of cards stuck in “ease hell”; they should come up if you search in the browser:

  • expression:*出禁[で*
  • expression:*天竺*
  • expression:*喘息*
  • expression:*燕返[つ*
  • expression:*巨木*

Thanks for reading! Cheers

2

I tried inputting your parameters into the FSRS simulator and indeed repeated Good answers on a card will result in rather slow interval growth; requiring 11 reps to reach 2 months when first interval is 11 days.

I also have a old and big language deck with a backlog where the optimized FSRS params just don’t seem to work well due to my review history being wonky. I’d suggest playing with the params in the simulator to make them do what looks reasonable to you and then updating them manually in Anki.

3

I was working on a more complete response, but I needed to jump in with this –

I strongly advise against you modifying your parameters from what the optimizer gives you. There are simply too many factors to consider for a mere human to generate parameters.

4

Before even looking at your data, there are 2 things we can get out of the way –

  1. It’s not “ease hell” – because there’s no such thing as “ease” in FSRS. Even if you subscribe to the idea that “ease hell” existed with SM-2, nothing that has ever been written about it applies to your situation now.

  2. If your FSRS parameters are well-optimized, on a substantial amount of accurate review history – there’s no reason to think that your intervals aren’t growing at the right rate for you – even if that isn’t the rate you think they should be growing at.

Yep, your parameters are tuned to grow your intervals slowly. The FSRS Visualizer gives you an easy way to view that, without needing to remember which parameter does what. You’re assuming that is bad or wrong though, and I don’t think that’s necessarily a reasonable assumption. It’s possible your intervals are growing just like they should be. If they were really growing too slowly for you, you’d be substantially outperforming your Desired Retention (DR) – but you’re not.

This is probably one reason. I’m not saying you need to change your grading habit, but you might need to accept this as a natural consequence of 2-button grading. You’ve given yourself no way to tell FSRS that some cards are harder and some cards are easier.

This is probably another reason. I took a look at the collection you have synced to AnkiWeb and you’ve got about 700 overdue cards out of 1200 active cards in this deck. That’s a lot. It means that FSRS is working with a lot of older review history, and it still doesn’t know how you’re going to do on those past-due cards. That suggests that as you catch up with current, you’ll be able to re-optimize and get some faster growth.

5

Hi everyone, thanks for the valuable insights :slight_smile: I feel like I understand the FSRS landscape a bit better now.

For now I will keep chipping away at my backlog, give 4 button grading a shot, and see how things go.

Cheers!

6

I might have just figured out the “issue” with my JP sentence mining deck!

By default, the FSRS optimizer excludes suspended cards:

image
image882×674 24.5 KB

However, on that deck I used to have a “retirement” addon which automatically retires (ie suspends) cards once they have an interval greater than 1 year.

The reasoning being, once the interval hits 1 year either:

A) You see that word frequently enough IRL that you don’t need the card anymore

B) The word is so uncommon IRL that it’s not worth spending time continuing to review it

That’s great and all, but it means FSRS isn’t analyzing about 85% of my deck, greatly skewing my review history / parameters! (Not to mention, of the remaining 15% of my deck, about half is currently overdue…)

I tried changing the search field to include my retired cards, and now my parameters look much more “normal”:

0.3734, 2.5251, 9.5300, 29.5433, 6.2134, 0.6328, 3.7241, 0.0521, 1.9055, 0.3554, 0.8715, 1.6324, 0.0592, 0.5339, 2.0768, 0.6014, 1.8729, 1.2047, 0.5737, 0.0100, 0.3265

preset:"Mined" -(is:suspended -tag:retired)

I haven’t tried doing reviews with the new parameters yet, but I’m cautiously optimistic