I’m a two button user with FSRS and I’m using Anki for language learning. In the past, my cards consisted of target language to native language translations of single words, and there I had good results with this approach. However, I’m now 7000 cards deep into a deck where I need to translate words into my target language and type them out, and I wish I had used the easy and hard buttons from the start, as very minor spelling errors count the same as having completely forgotten the word. Not only that, but a good chunk of the words are very easy and those count the same as words that take some effort to remember. I know that statistically using 2 buttons seems to work better, but I’m skeptical if that’s also the case for my deck.
Anyway, what I noticed is that if I exclude all words that I’ve never answered incorrectly and that I have repeated at least 3 times and that have an interval of more than 21 days, the RMSE jumps from ~3% to ~12%, and to ~7% after an optimization. To me that seems to indicate that the parameters are very much skewed towards the easiest cards, while the harder cards are actually the important ones. I’m now contemplating if it makes sense to just periodically suspend the easiest cards, so that the harder cards get repeated more frequently and more closely adhere to my configured retention rate.
As for the buttons: I think with the number of reviews that I’ve done, it wouldn’t make sense to start using 4 buttons now, as the meaning of “again” will always be contaminated.
Any thoughts on this?