I have noticed that, with no new cards in the deck, the number of reviews is not coming down. So, I have started to pay attention to the intervals and noticed that sometimes they go down even if I answer ‘good’. I understand that this could happen if you change parameters of the deck. For example, if you optimize FSRS parameters, but this is happening even when there’re no parameter changes.
Below is one example of this happening. I do parameter optimization once a month and you could see it has been done last time on 1st of December due to ‘Rescheduled’ record (I check the ‘Reschedule’ option on parameter optimization). No other parameter changes has been done either. Yet, you could see interval going down on 15th and 29th after answering ‘Good’. What could be causing it?
Good idea, I bet my observed behavior is caused by fuzz. In the back of my mind, I had thought it was something exotic such as: “even though he remembered this card after 8 days, he is likely to forget it next time after only 7 days.”
I don’t think that’s wise. You should grade your answers honestly and accurately – and without regard for what the next interval will be.
From your review history, it looks like the card holds steady at a certain Stability for a while – through a number of reviews and weeks. If S is at about 4 days, and it’s not increasing fast enough to get to 5d after several successful reviews, whether you study the card after 3 or 4 or 5 days is probably a distinction without a difference. So don’t worry so much about the specific interval the card gets.
This is not because of fuzz. FSRS considers Hard to be a passing grade. So, it increases the interval when you press Hard. Whether it becomes double or 1.5 times or something else depends on your FSRS parameters. Your FSRS parameters may be bad if you have a history of inconsistent grading, which your Forums comment suggests (I even added the current interval to the back template so I could factor it into my answers.)
Anyways, triage is better done in the Forums than here. You may share your FSRS parameters and review history (collection file) there so that someone can try to identify the cause.
I only did it a week ago. But this has been going for a while. I’d think after more than a year the algorithm should have been able to adapt…
Every time I optimize the parameters it either says I already have optimal ones or that it could adapt well but it still increases intervals way too fast too often.
Come to think about it. Does the FSRS try to keep passing rate at 90%? I.e. it would be lowering intervals if the rate is below 90% and raising them if it is above?
If so that might be my problem… For the decks that do not have any new cards maybe I need to start gradually raising the target pass rate.
FSRS schedules each card for when it calculates Retrievability for that card will pass your Desired Retention (DR). Is your DR set to 90%? Then that’s when you cards are being scheduled, but the due date isn’t continually changing. Once it’s set, it stays put unless you reschedule.
(And by the way, don’t use reschedule-on-change unless you have to. It adds an entry to the review log for every one of your active cards. That’s really unnecessary.)
Some of the examples you’re posting in GitHub seem like they could be explained by re-optimization between those reviews. If you have more of those in the future, they would be better posted here. But you’ll also need to give us your parameters as text, please, and DR for the time when that review was done. If someone needs to look further into your collection, they’ll ask you for that.
Does this also mean that if I reset cards the old pre-reset reviews are being counted towards setting new intervals? (This is about a different deck than above)
