Observations after a while of FSRS

Anki FSRS
1

After some weeks of FSRS I find a behaviour that needs getting used to.

I have a deck with 90% desired retention.
Two phenomena appear:

  1. quite often, a card keeps the same interval a number of instances in a row, and only then jumps up. (see here four times 2 days, and two times 4 days.
  2. Sometimes, the Interval even goes down after a “Good” answer - see in the example where it goes from 4 to 3.

image
image513×338 6.2 KB

I don’t have a specific question, I just find the behaviour curious. The repetitions even seem to make some sense, as it happens that I fail to answer correctly even in the third repetition. Maybe someone can explain how this works, and what makes FSRS reduce the interval from 4 to 3 after two successes.

1 Like
2

That can happen if you have re-optimized parameters between these repetitions. It can’t happen with the same parameters.

4 Likes
3

What you’re seeing is based on your parameters and the particular history of this card.

  1. This has to do with how fast the Difficulty of a card drops with each successful review, and how fast the Stability rises with each successful review. Just outside of your screenshot is the D calculated at each of these reviews. Even without seeing this card’s complete review history, I expect it’s quite high.
  2. As Expertium said – different parameters (or different versions of FSRS) between one review and another can cause this. Unlike SM-2 FSRS does not consider the length of the current interval when setting the next interval. It is starting from the card’s current memory state (D, S, and R), which is recalculated when you update your parameters.
4 Likes
4

It does, since R depends on the interval length

2 Likes
5

Where? I thought the only variables in the R function were S Stability and t time since the last review. A technical explanation of FSRS | Expertium’s Blog

6

Yes. Then what do you refer to when you say “length of the current interval”? Do you mean different things when you say “current interval” and “time since the last review”?

7

Ah, yes, the interval is the assigned time delay – that thing in the “Interval” column. Under SM-2, it was a primary factor in determining the next interval.

But the “time since last review” is an actual passage of time, which could be longer or shorter than the interval. With FSRS, as far as I can tell, intervals are used to add days and set a next due date, but no actual computation is ever done with them. [Which is why in Set Due Date, changing the interval doesn’t matter anymore.]