I recently switched to FSRS and let it reschedule my cards. I thought I understood what some of the terms mean, but after looking at this card’s stats below, I’m no longer certain.
I set my desired retention to 85%. The interval after my latest review is 2.4 months, but according to the projected forgetting curve, 2.4 months puts me at 69% retrievability.
Also, the stability is 1.4 months (and I’m not sure what the (43) means), but the projected forgetting curve says 1.4 months puts me at 77% retrievability, not 90%.
The only thing that seems not to match is the “projected forgetting curve” – which looks like it drops much more sharply than the historical curve has through the last few reviews, doesn’t it? If you set aside that visual, everything else makes sense.
Some things to look at –
I see a review history that looks like you kept your cards in Learn/Relearn for a long time when you were using SM-2 – and now it looks like you possibly aren’t using learning/relearning steps at all – is that right? FSRS doesn’t have a model for near-term memory – so blanking out your steps isn’t recommended for most users – Reddit - The heart of the internet .
I assume you just optimized your parameters for this preset – did FSRS count a lot of reviews? [If you don’t remember, you can click optimize (current preset) to see what it reports, and then close without saving.]
If you’re using different versions of Anki on different devices, they might have different versions of FSRS. Which one did you optimize your presets on? The other might not be able to use the same parameters. So if you studied the card and it was scheduled on one device, it’s possible that the “predicted” part of the graph on the other device is using different data. [This is a new theory I’m having right this moment – and I’m not entirely sure what goes into producing the graph, but given where things like parameters are stored, it seems possible.]
EDIT: SCRATCH ALL OF THE BELOW. The issues (different graph, different intervals) were seemingly solved by forcing a one-way sync on my Surface, making it download from AnkiWeb. All devices currently show the same graphs, and identical reviews give identical intervals. I did force that one-way sync before, I believe, so I’m not sure how things fell out of sync.
That’s what I originally thought, but I just did a few identical reviews on my 3 devices (explained below), and the intervals from my Surface were significantly longer than my Desktop and iPhone! So the underlying calculations do seem to be affected.
To your other points:
I do still have learning/relearning steps. I used to use long Learn/Relearn steps in SM-2, but when I enabled FSRS, I tuned those both down to 30m 5h based on the docs saying “Ensure that all your learning and re-learning steps are shorter than 1d and can be completed on the same day”.
Yes, FSRS counts about 90k reviews. I only have 1 preset (for now), though I know the docs recommend different presets (and thus parameters) for different kinds of decks.
I have 3 devices. Desktop [Version 25.07.5 (7172b2d2)], Windows Surface [Version 25.02.6 (6381f184)], and iPhone [Version 25.09]. I don’t remember which I optimized parameters on (either Desktop to Surface), but they do all list the same parameters in Options. The above image is from the Surface (25.02).
As an extra little wrinkle, the graph does look different between Desktop and Surface, in line with some version mismatch issue. BUT I updated to 25.09 on both Desktop and Surface, and the graph changed on Surface but not Desktop. Both shown below. Toggling off add-ons had no impact.
tl;dr 2 Windows machines which currently have the same version of Anki but yield different intervals after reviews.
