i’ll have a look
Btw, once the next major version of Anki comes out, I will help AnKing to make another video about FSRS and stuff. I have a draft where I outlined all the major changes, here’s a short summary:
Anki 23.10
• FSRS v4.
Anki 23.12
• FSRS-4.5. The number of parameters is the same, but the shape of the forgetting curve has been changed.
• “Optimize all presets” added.
Anki 24.04
• Compute optimal retention → Compute minimum recommended retention (I will call it CMRR from now on, because the full name is a bit of a mouthful). Details: The Optimal Retention · open-spaced-repetition/fsrs4anki Wiki · GitHub.
• “Ignore reviews before” [date] added. Details: fsrs4anki/docs/tutorial.md at main · open-spaced-repetition/fsrs4anki · GitHub.
• Fixed a major bug – parameters being worse after optimization than before.
• The minimum number of reviews required to run the optimizer was decreased from 1000 to 400.
Anki 24.06
• The optimizer no longer requires a minimum number of reviews. Instead, it uses a complex rule to decide between 3 actions: full optimization, “pretrain” or using the default parameters. “Pretrain” is when only the first 4 parameters are changed, the rest are set to default. It’s useful when the number of reviews is too low for full optimization, but not 0. Which option is used depends not only on the number of reviews, but also on what kind of reviews those are.
Anki 24.XX
• FSRS-5. It has 2 more parameters and finally takes into account same-day reviews.
• CMRR is also more accurate thanks to taking same-day reviews into account. Also, it no longer requires a minimum number of reviews, and has its own default parameters. So now, both FSRS and CMRR can be used by anyone with any number of reviews. Additionally, CMRR is now more accurate than before when “Days to simulate” is <365.
• Load Balancer. It’s not actually a part of FSRS, and it works with both SM-2 and FSRS. Load Balancer is like Smart Fuzz or Fuzz 2.0. It slightly changes intervals to make sure that the number of due cards is very similar across all days.
• Easy Days. Again, it’s not a part of FSRS, and works with both SM-2 and FSRS. It allows users to select days of the week, as well as specific dates, when they do not wish to study (or wish to study less).
• New FSRS-related stats.
• Ignore reviews before → Ignore cards reviewed before. It doesn’t actually ignore reviews, but entire cards, the name was misleading.
This is missing minor changes, such as “changed the range of this one parameter” or “tweaked CMRR for 3908506th time”.
2 Likes
24.xx is current anki main branch, i’m assuming?
It’s a future version. Idk when it will come out
1 Like
The minor changes are recorded here:
2 Likes
What is the chance that FSRS could begin scheduling <1d intervals automatically in the future, provided that it will start taking learning steps into account 
2 Likes
Very low. @dae if you don’t mind, please give a thorough explanation about all the database stuff and why learning steps can’t just be removed
No, I am not talking about removing learning steps altogether. Just that review intervals can become <1d
“Letting FSRS schedule intervals <1d” and “Completely changing everything about how learning steps work” are practically the same thing.
Oke…so I suppose having a fixated set of learning steps (and relearning steps), for example, 10m 1h, as usual, would collide with FSRS scheduling a review in 6h. I mean I could get that could lead to some problems perhaps if a learning step is greater than a review interval (for example, if FSRS would schedule a review for a card in 3h whereas as person has a learning step interval of 6h)
I don’t think learning steps will need to change but how review cards works need to change. Currently, we show cards after a day rollover so that you don’t have reviews popping up every 5-6 hours in a day.
I personally think longer sub-day intervals are not worth it, and wouldn’t want FSRS to be telling the user to wait a few hours to review again.
: (
How about a hard cap
If a user can’t manage to review the subday interval, it can always be deferred to the next day.
I mean this feature is already present in some form
I purposely have very long sub-day learning steps (about 6h) and I study till past midnight. So if the subday interval is longer than the time left till the hard-cap it is automatically sent to the next day.
1 Like
If it needs adding extra complexity to the app in the form of new options I am against this. It’s also much less intuitive than learning steps to have a max limit on delays determined by FSRS. I wonder how will we explain how the learning phase works in manual.
I mean the feature is already present right there. I myself wasn’t aware of it until 4 months ago and I myself have been using Anki for 2 years 
So I suggest just adapting this feature if FSRS scheduling subday intervals is inconvenient to some.
I mean, another step towards FSRS cram mode is a thumbs up in my eyes.
However… I would like to see all of this done in a proper way together with an onboarding revamp.
2 Likes
Yes indeed 
FSRS already has a cram mode, it’s called 99% desired retention.
In a way that actually makes stuff like stats and stuff intelligible, is what I mean.