I recently migrated to FSRS for new cards and my new cards that I have pressed again on once (never seen before that point) get scheduled 2 days out. I see a lot of people on here that have this issue when pressing hard or good and I understand that that is FSRS being optimal yet counterintuitive. Is this the case here as well? 2 days for pressing again with no prior exposure seems like bug to me. Is it because FSRS still has to learn about my behaviour concerning new cards?
These are my FSRS parameters: 1.7442, 9.1067, 24.3739, 24.7067, 7.0431, 0.6677, 1.4555, 0.0878, 1.8121, 0.0000, 1.2359, 1.9766, 0.1046, 0.3938, 2.3419, 0.1912, 2.9459, 0.6020, 0.6395
You probably have empty learning steps. If that’s the case, adding 10m
or something like that should fix it.
Yes, I have empty learning steps. But as I understood from this Anking video (THE ULTIMATE 2025 ANKI SETTINGS — Latest Updates, FSRS-5 & More!), you should now be able to leave learning steps blank and FSRS will schedule the cards for you immediately. However, I expected it to schedule in minutes or hours, not days after the first exposure and pressing again.
- You can leave them empty, but FSRS doesn’t have a short-term memory model, so it’s not going to work well when it comes to same-day reviews.
- Personally, I suspect that it’s actually optimal to NOT have same-day reviews in almost all cases
Ok, thank you! I expected at least the “again” to be same-day, but maybe I just haven’t understood FSRS well enough yet.
I also extremely strongly suspect this, both for learning and for relearning. Even with FSRS’s lack of a short-term model, it’s probably still better than literally any other number or (readily available) algorithm one could put in there or whatever number any given user would put in. It might be worth it to set the default to ““ instead of 10m.
In one extreme case, in the case of me re-memorizing some information that I previously memorized 10 years ago and have since forgotten, FSRS has decided that my initial “Again” should lead to a 16day interval. It’s… more or less correct.
I have been using it with FSRS with no learning steps for a while now and the 2 days interval turned to be because of hard misuse. Fixing that I don’t always get long intervals on Again.