How the review forecast works

I’m curious how the review forecast works. Naturally, it doesn’t know yet what I’m going to do in terms of working through reviews or adding new clozes between now and the times for which it makes predictions. So does it base its predictions only on the reviews that are already “in the pipeline”, or does it have a component that is based on my previous behavior?

The most important question to me is to which degree I can trust the forecast, at least in the near term. If I felt that I understood better how it worked, I’d feel more free to put off reviews on a day when I have a lot of them, being relatively sure that I could go through them on a day when I have only a small number scheduled. As it is, every day I clear out my entire review queue and do five rounds of new sentences. It would be nice to feel as though I have a little more control without worrying that the number of reviews will become unmanageable.


I’ve wondered the same and studied its behaviour a lot previously. I don’t know if it’s changed, but in my previous investigations, I found that it only will ever show the actually scheduled reviews, for the dates they’re due. If you “play” more sentences, you’ll see that they’ll get added on in the appropriate places in the forecast according to their corresponding % mastered interval. There doesn’t seem to be any intelligent forecasting based on “normal” playing/reviewing behaviour as far as I could tell previously.

I would really like at least some sort of basic distribution option, where you could play around with “I would like to play x sentences per day and review y per day” and see how the review forecast would be changed accordingly (ideally while also taking your review preferences of “oldest/least mastered/… first” into account), since that would really help me play around with developing a good balance between the two, rather than just being confronted with heaps of reviews to clear, then having nothing, so just overdoing it again with playing new sentences, etc. I tried to make my own predictions, but since there’s nowhere that you can easily read in the currently scheduled upcoming reviews, and I’ve played around with the review intervals quite a bit too, I didn’t immediately manage to come up with anything usable.


Many thanks for the thorough analysis, @sindaco. It’s helpful to know that past behavior is ignored in the forecast. If there is to be only one kind of forecast, I think it’s better to have it work that way.

Like you, I would prefer to make the number of reviews per day more constant, either via a setting that would set it directly, or (less ideally) via tools that would help us determine it for ourselves and then set it manually. In Anki (a general spaced-repetition learning tool, for those of you who are not familiar with it), you can specify the maximum number of reviews per day. In fact, the interface forces you to do it (though setting it to a very high level means that the number of reviews will effectively be controlled by the number of reviewable sentences for the day rather than the limit you specify).


This is correct - thanks @sindaco! We’ll aim to add an explanation along these lines to the UI.

I’ll look at how Anki does it as well, thanks for mentioning, but how are you envisioning it might work? You’d set the max reviews per day to 50, and anything over that gets bumped to the next day, and anything over for that following day gets bumped to the next day, and so on? So for example if you had 100 due today, 75 due tomorrow (today + 1), and 50 due the day after (today + 2), and you were to look at the review forecast the way it currently works, it’d show 50 per day for today, 50 for today + 1, 50 for today + 2, 50 for today + 3, 50 for today + 4, and 25 for today + 5 (apologies if this notation is difficult to understand, but hopefully it conveys the gist)?


Yes, that’s how I envision it. Thanks!