I think it would be better to split the DR for young and mature cards.
Young cards are less stable, and I would prefer to use a lower DR for them.
Meanwhile, mature cards are more stable and easier to review, so a higher DR would be better for them.
I think a lot of users would not like to re-learn old cards with 1-year intervals, but they would love to speed up interval growth for young cards.
Hi,
Thank you for your reply and advice!
What I meant is that cards still in the learning phase (i.e., young cards) and those already learned with mature intervals should have different DR values, in my opinion.
If I simply increase the DR for all cards, it ends up giving me more reviews—most of which come from young cards.
It’s like setting a higher DR for mature cards results in a kind of “punishment” through the increased review load caused by young cards.
Theoretically, I could set a 90–95% DR for cards with ~1-year intervals without getting too many reviews. But I can’t do that if it also affects younger cards with intervals of just a few weeks or under 30 days.
Therefore, I think it would be more efficient to split the DR between cards that are already learned and have long intervals, and those that are still young and fragile. Increasing the DR for mature cards wouldn’t cause an avalanche of reviews, while doing so for young cards would result in too much workload.
I think many users would prefer to use FSRS not just as a prediction tool, but also as a system to improve long-term retention. It doesn’t seem fair to apply the same DR to both 1-week and 1-year intervals.
Personally, I’m okay with forgetting 10–30% of young cards. But I really don’t want to forget 10–20% of cards I’ve already learned and which now have large intervals. That would mean having to re-learn 10% of my mature cards after a year—cards I thought I already knew.
I think 95% DR is brutal on young cards. And at the same time lowering DR globally to 85–90% sacrifices mature cards you already knew as with that I agree to forget 10% of cards with ~1 year interval.
Alternatively, instead of manually setting two separate DR values for young and mature cards, I suggest configuring FSRS so that the DR increases over time as the card’s interval grows.
In other words, the longer the interval becomes, the higher the DR should be.
For example, DR could increase by a fixed amount (say, N) every 30 days of interval growth or based on some smooth function. Ideally, the user would be able to configure the rate of increase (i.e., how much DR should grow per time or interval unit).
This way, DR wouldn’t be a fixed value but a dynamic one that scales with the card’s maturity. That would allow younger cards to be reviewed with a more forgiving, lower DR where forgetting is acceptable and older, mature cards to benefit from a higher DR to prevent unnecessary re-learning.
This has been brought up before. I guess the only issue here is complexity, otherwise it’s something that can be of help.
Thank you for your reply!
I didn’t find any discussion like this on the Anki Forum, so I thought I could publish the suggestion, since it’s not a duplicate.
Could you please explain why it’s difficult to have different DR values for young and mature cards? As a user, I don’t really understand how that would be fundamentally different from setting a single DR for all cards. Sorry if that’s bothering you.
Jarrett mentioned that it’s possible to achieve this by focusing on optimum intervals: [Feature Request] Separate retention rates for “learning” and “known” items · Issue #694 · open-spaced-repetition/fsrs4anki · GitHub
But why is FSRS focused on DR instead of OI, then? Is there a reason for that?
I’m just trying to understand this. Because separating cards with short (e.g., week-long) and long (e.g., year-long) intervals using different DR values seems like a logical next step in the evolution of FSRS. But if I’m missing something, could someone explain it?
If I understood correctly, the reason FSRS doesn’t cover the learning steps is that there’s no proper model for short-term memory, as FSRS was designed specifically for long-term retention.
But when it comes to splitting DR between young and mature cards (or something like this), we do have a working model. So what’s the reason for not implementing it? I’m just trying to understand it.
See:
But why is FSRS focused on DR instead of OI,
Example logic explaining DR vs intervals vs ultimate goal of using Anki:
Conclusion: ultimate goal is to pass exam (DR). Anki achieves this for young and mature cards by adjusting intervals of the cards. So DR should be consistent for all cards but intervals should be adjusted per card.
For the challenge you can put: “I want to impress in front of my colleagues”, “Win the contest”, “Use knowledge at work”. If you use Anki just for using or to feel better (you have no goal) then I can understand why DR could be confusing.
I appreciate the explanation, but I think you’re misunderstanding the core of my suggestion. I’m not confused about DR or my learning goals.
My suggestion is a refinement of FSRS behavior to allow DR to scale with card maturity, because memory stability differs between young and mature cards.
The goal is still the same: reaching a certain retention level. But the cost structure of reviews varies drastically between early and late stages of learning.
Uniform DR might be easy to implement, but it’s not necessarily optimal.
If you’re only using Anki short-term for a test, maybe this doesn’t matter. But for long-term use, DR dynamics could significantly affect review load efficiency.
Though not exactly what you requested for, Jarrett is developing a feature that will allow you to set different DRs for different decks having the same preset. Then, you can move your mature cards to a new deck and set a different DR for that deck.
Thank you for your reply!
That’s great info. I’ll make sure to use the feature once it’s available.
It is possible. If I understand you now right you try to tell me that there is proof that FSRS work incorrectly for mature cards - it gives more load than needed. If you were right then this could be seen in long term statistics. Do you have such statistics? Can you show me base on what data you claim that there is problem?
I have in all decks 90% DR (I use separate config per deck type for better FSRS values), since I use FSRSv6 all statistics are adjusting to this value
If you were right I should see much higher retention for mature cards in last month (because I would see much often mature cards then it is needed like you said).
FSRS exactly match my memory and total for young and mature is around 90%
@zZH2z, you are still misunderstanding.
Jarrett’s research has shown that having a low DR early on and a progressively higher DR as the stability increases is more efficient for learning. More efficient in the sense that it requires less number of reviews for similar learning outcomes.
The main issue is that it’s difficult to calculate the matrix of optimal intervals.
Here’s the relevant repo, but it’s not being actively worked upon: GitHub - open-spaced-repetition/SSP-MMC-FSRS: Stochastic-Shortest-Path-Minimize-Memorization-Cost for FSRS
Thanks for your follow-up.
“If I understand you now right you try to tell me that there is proof that FSRS works incorrectly for mature cards – it gives more load than needed.”
“If you were right then this could be seen in long-term statistics. Do you have such statistics? Can you show me what data you base this claim on?”
I’m not claiming that FSRS has a “problem” in its current form. What I’m saying is that there’s an obvious next-step improvement that could make the system more efficient by giving young and mature cards different treatment through separate DR values.
Thank you!
Is there a real technical reason why DR hasn’t been split between young and mature cards, or is it just a matter of avoiding complexity?
Do you mean allowing the user to set one value for young cards and one value for mature cards in the Deck Options page?
I think the main reason why this wasn’t (and won’t be) done is that it would make configuring FSRS much more confusing for new users.
Do you mean allowing the user to set one value for young cards and one value for mature cards in the Deck Options page?
I think the main reason why this wasn’t (and won’t be) done is that it would make configuring FSRS much more confusing for new users.
It feels a bit unfair that users who’ve taken the time to read the manuals, understand FSRS, and commit seriously to their studies are held back from using more powerful features just to keep things simpler for beginners or average users.
Actually, it can be. There’s an advanced section at the bottom. A new option can be added there.
However, there is one more important issue. How do you define mature cards? Conventionally, Anki used the 21d interval to distinguish them. But, if we alter the DR, the interval changes too. Secondly, 21d is not a good cutoff for this use-case, IMO. It should be something larger like 6 months or 1 year.
Also, the separate deck approach I mentioned above gives more control to the user. So, that’s probably better, even though it’s more cumbersome.
Use stability.
If this is effective, then for a start it could be implemented using FSRS Helper. If it becomes popular, there is a high probability that the feature will be added to anki.
