I’m not asking which approach is better. I’d just like to understand the difference between these two scenarios from the perspective of the FSRS algorithm.
Scenario A
- I use FSRS 6.
- Learning steps =
10m.
- Relearning steps =
10m.
- I use Anki this way for one year.
- Then FSRS 7 is released, I update, re-optimize my parameters, and remove the learning/relearning steps if appropriate.
Scenario B
- I use FSRS 6.
- Learning steps are empty.
- Relearning steps are empty.
- I use Anki this way for one year.
- Then FSRS 7 is released, I update, re-optimize my parameters.
From the algorithm’s point of view, what is the difference between these two histories?
I’m not asking which scenario is preferable. I’d simply like to understand what information is different between them, what each history contains that the other doesn’t, and how a future version of FSRS would view these two users differently, if at all.
They should contain the same thing. The difference is that FSRS decides what intervals are being used short term if (re)learning steps are left empty (see Deck Options - Anki Manual), whereas if you set your own (re)learning steps, it’s you that controls that.
Given everything is identical (same cards, same learning habits, same memory, ect.) and the only thing being different is whether you set your (re)learning steps yourself or let FSRS do it, the result would depend on whose values where better.
If FSRS did a better job at the short term intervals (compared to those steps you would manually set), then your workload with empty learning steps is lower, else higher.
Same applies after reoptimizing.
You’re asking how FSRS 7 will view those review histories. But “FSRS 7” hasn’t been fully implemented yet, and hasn’t been submitted as a pull request. [I welcome correction if I’m not up to date on that.] I understand that it’s being worked on somewhere, but the people working on it are the only ones who know what it might eventually consider and how it eventually will work when it is submitted (and even they don’t know whether all aspects of it will be accepted and merged into Anki).
This seems like a hypothetical question about a hypothetical algorithm, so I’m not sure it’s possible to answer.
If you want an answer about how FSRS 6 would view 2 review histories that differ in the number and/or length of re/learning steps used – that’s still hard to answer, but let’s take a shot.
First, there’s no one-size-fits-all answer, because your optimized parameters are completely different from someone else’s (and from the defaults). You can’t assume as a starting point that the only difference between these review histories will be in Learn/Relearn.
Depending on your parameters, the number of successful grades in Learn/Relearn might be a significant factor in scheduling later on. And since you trusted your steps to an algorithm that didn’t have a model to determine them accurately, they may or may not have been helpful to your learning process. So there could be huge differences between these 2 review histories.
But FSRS doesn’t care much about how or why or when cards were originally scheduled. When it looks back at review history – during optimization, or when scheduling a single card – it compares the review date-times and outcomes using your current parameters, and what the calculated Retrievability would have been at those points.
So if you’re asking something along the lines of – “does it matter to FSRS [6] whether I studied with my own learning steps or with its learning steps?” – it does not. But if the question is more like – “does it matter to FSRS [6] when and how many times I studied a card and what grade I gave it?” – yes, it does.