The last 0.5% #228
Description
FSRS-6 with recency weighting still underperforms Anki-SM-2 with default parameters, which is neither optimizable nor even designed to predict p(recall).
I think we may learn something interesting by manually analyzing those 50 or so users.
@L-M-Sherlock when you have some free time, please send me a dataset with only users for which FSRS-6-recency performs worse than Anki-SM-2 with def. params. Though maybe I'll do it myself.
Maybe those users just mindlessly spam Good and therefore have like a 99.99% retention or something. But if not, we may learn something interesting about edge cases.