Ordering Request: Reverse Relative Overdueness

Not only should Reverse Relative Overdueness be available in Anki, but I believe it should be the default sorting method for most decks and situations, especially now that we have FSRS. I’m not pushing for it to be the default right now—just making it available would be great.

Side note: As I was finishing this post, I realized that the optimal sort option would actually be based on the derivative of the forgetting curve—that is, prioritizing the cards you’re going to lose the most retrievability on if you don’t study them today. If anyone knows how to implement that, please consider this post obsolete and go with that instead. Anyway…

I requested something similar years ago:
https://forums.ankiweb.net/t/another-choice-for-order-in-filtered-decks/3255

And I noticed it’s been brought up in another recent post (see point number 9):
https://forums.ankiweb.net/t/several-fsrs-related-suggestions/44673

I also searched and saw this has been requested multiple times, which doesn’t surprise me. It’s a glaring missing feature. I’ll argue why it should probably be the default sorting method, just to emphasize the importance of simply having it available.

When cards come due, you want to see them at the sweet spot—right when their retrievability dips below the desired retention you’ve set. Some people worry that new cards will constantly pile up in front of other due cards, but that’s not the concern they think it is.

At the sweet spot, cards are at the steepest part of the forgetting curve. They’re going to lose retrievability faster than the cards they’re piling on top of, especially if their stability is low. Because of that, there won’t be as many of them. Cards not studied right away quickly drop from higher retrievability levels and settle into backlogged blobs as they wait longer. You want to study those cards before they rapidly lose retrievability, then work your way through the next ones that haven’t lost as much, effectively working down the retrievability scores.

The default advice for working through a backlog (https://docs.ankiweb.net/filtered-decks.html#catching-up) works okay if you’re studying consistently every week, but you have to set up custom decks, which isn’t necessary with my suggestion. Plus, if you don’t study for a week, you’ve lost cards to the morass. Another suggestion I’ve seen on reddit (https://old.reddit.com/r/Anki/comments/1b40ah5/my_exams_are_in_two_months_im_currently_focusing/ksxsd1u/) is even worse. If you skip a single day of studying, those cards are lost to the backlog and will have lost a ton of retrievability by the time you get to them.

With Reverse Relative Overdueness, you don’t need to create custom decks or worry about how long you’ve gone without studying. The low-stability cards that came due during your absence will be at the bottom of the pile—where they belong because you’ve forgotten those the most.

After a hiatus, as you study day after day, cards you get wrong and that come back up again will show up at the top of the pile, ensuring they stay maintained as you work through your backlog. It’s exactly what you want when you’re using spaced repetition.

3 Likes

The closest alternative that people use when there is a large backlog is sorting by decreasing intervals.
The proposed sorting will better match the users’ wishes.

I’m not sure if it’s worth doing it according to the default settings. Will such sorting show the cards randomly enough in the absence of a backlog?

Why would it be any less random than the current default? It also depends on how many significant figures Retrievability is calculated to. If it’s several digits, then it’d be very much random from the user’s perspective. If it’s like two digits, then there will be many ties and tie breakers could be handled in the same way that the current default uses.

If there’s no backlog, they’ll all just be showing up on their due date. If there’s a small backlog, the cards that showed up today will be studied first, which is what you should want.

Also, don’t sleep on the side note. I’ll be screaming this from the roof tops in about a year or so (unless I’m told it just isn’t feasible, then I might try to figure it out myself). This is really the ultimate answer.

But for now, Reverse Relative Overdueness is a big step forward.

Retrievability drops faster for cards with less stability. For some cards, stability may not even reach a day. Cards with low stability are more likely to be shown last.

Only if you skip a day. That first day they’re due, they’ll be right at the top with all the other stability scores. And if you skip one day, they won’t get piled on too much unless you aren’t studying enough to work your way through the current days cards. But if that’s the case, then you just aren’t studying enough to keep up anyway and you’re going to be building your backlog regardless of what sort you’re using.

I think most people are at least going to find it easy to get through the current days due cards, and then those low stability cards you skipped yesterday will be the next ones up.

I’m talking about a situation where we don’t have a backlog.

If there’s no backlog, then cards with low stability are not more likely to be shown last. Stability won’t factor in at all.

I just want to say, I heavily edited the part of the manual documenting all the sort orders. And yet, my head spins everytime I look at them. I’d want to slash some of those rather add more. Net-net, we’ll be gaining users if we get rid of the mind-numbing plethora of options designed for all kinds of Anki fetishes.

Edit: Urgh, I am unable to link to the UXMYTH article about why reducing options leads to greater user satisfaction. But anyway, if we add more options, I also demand a order that prioritises cards of even R values over odd R values and this order reverses every week.

Totally agree with getting rid of some. I think this is one of the most important ones though. Get rid of a bunch and add this one. Again, it probably should be the default. It’s at that level.

LMSherlock analyzed different sort orders. TLDR:

  1. If the user can do literally any number of reviews per day, the order doesn’t matter
  2. If the user can only do a limited number of reviews per day, then Difficulty Ascending is the best in terms of spending the least amount of time studying. Recommend sorting by relative overdueness.md by Expertium · Pull Request #634 · open-spaced-repetition/fsrs4anki · GitHub. Retrievability Descending is the second best, which is kind of what Reverse Relative Overdueness would be.
1 Like

This might be the opposite actually. Sure those decreasing intervals in the 0.8 to 0.9 range will show up earlier, but when you have a large backlog and start working through it, this would never allow cards you’re getting wrong today and tomorrow to show up the next time they’re due. They’d always be at the bottom next time up.

I don’t trust it. Are you saying sufficient interleaving with a random sort order doesn’t aid in memory (I haven’t read whatever u linked us to).

It’s exactly Retrievability Descending if the cards’ desired retention level is the same.

I’m skeptical, and considering the formula for Difficulty is the least interpretable (or most arbitrary) of all of the metrics, probably not a good idea to actually use that sort.

I could see that making sense for going through lower difficulties first, but the vast vast majority of my cards are in the 95% - 100% range, and I doubt the ordering within that space is helping much.

Generally, you don’t trust an idea fresh out of the oven. (Actually, when did this “reaserch” happened? Today?)

The stuff on Github that I linked? In April

We could add Reverse Relative Overdueness and remove a few other ones, especially the ones that are worse than random, but I’m afraid that for every sort order there will be someone screaming “No, I use it all the time!”. Also Dae won’t care

I’m talking about Expertium. I haven’t even read your post. I already know I wouldn’t want a new sort order.