This post replaces my old post about benchmarking and I added it to my compendium of posts/articles about FSRS. You do not need to read the old post, and I will not link it anywhere anymore. First of all, every "honest" spaced repetition algorithm must be able to predict the probability of recalling a card at a given point in time, given the card's review history. Let's call that R. If a "dishonest" algorithm doesn't calculate probabilities and just outputs an interval, it's still possible to co...