FSRS shrinks the interval upon "Good"

  1. On 2024-01-24, I studied a card and answered “Good”. The legacy SM2 scheduler assigned an interval of 3.77⁩ months.
  2. I switched to FSRS two weeks ago.
  3. On 2024-05-16, I studied the same card and answered “Good”. The FSRS scheduler assigned an interval of ⁨1.97⁩ months.

Note that the card’s interval decreased even though I answered “Good”. This seems a bit counterintuitive to me.

My desired retention is 85%. So it seems like FSRS is telling me: “Even though you remembered this card after 3.77 months, you have less than 85% chance of remembering it in 3.77 months.”

FSRS may well be correct—but is this behavior expected?

Yep, expected.

FSRS determines the “memory state” (current Difficulty, Stability, Retrievability) for each card the first time it sees it. To do that, as I understand it, FSRS looks at the entire review history of the card, and calculates D-S-R for every day you’ve reviewed it. But FSRS doesn’t pay attention to what the SM-2 algorithm has been doing with the card – it makes its own decision at each step about the new memory state, with fresh eyes.

So on 2024-05-16 when FSRS first saw that card, it disagreed with 3.77m being the correct interval. (Actually, I bet FSRS consider the card to be already overdue, so it would have given you a “bonus” for getting it correct after so long.) To see why FSRS did that, you’d have to look at the whole history of the card.

This is the same phenomenon that leads the the recommendation against “reschedule all” when you’re turning FSRS on. If you’d done that, this card would have been rescheduled to a much earlier due date. By letting FSRS see your cards one-by-one as you study them, you avoid every card in your collection being suddenly shifted forward or backward at once (the huge-backlog problem) – but the cards still are subject to being shifted to longer or shorter intervals as they are reviewed.

Because of the way FSRS works, there will always be outliers – cards that don’t behave how most of your cards behave. Have you optimized your FSRS parameters or are you using the defaults? Either way is fine, of course.

2 Likes

Yes, I have optimized the FSRS parameters. My desired retention is higher than my historical retention, so I am fine with getting some shorter intervals. Also, I didn’t reschedule my cards—it seems better to shift to FSRS schedules gradually.

So far, so good after two weeks; I will be interested to see how my retention changes after some more use.

Not rescheduling when you start is definitely the way to go. And yes, keep an eye on your retention. That’s the best yardstick for how FSRS is working for you.

If you have a lot of Deck Options presets – the switch turns FSRS on for your whole collection, but each preset has its own parameters and desired retention. You might want to check that you’ve got your desired retention set correctly for every preset.