Maturity state is currently based on interval. This leads to changes in desired retention having a significant impact on maturity counts, which seems counter intuitive.
My intuition of maturity is classifying cards based on how well I remember them, ie the stability of my memory of them.
Does it make sense to redefine maturity to be classified based on card stability?
That’s a good idea, but it would cause inconsistencies when switching from SM-2 to FSRS or vice versa
That doesn’t change the fact that mature is arbitrary category. Any ideas on that?
You could also see it as reducing inconsistency, as FSRS with non-90% target retention introduces very large inconsistencies in mature counts.
I don’t think the arbitrariness of 21 days is an issue. It’s a nice at a glance metric for how many cards I remember “well”, and how stable my level of knowledge is. I prefer it over the estimated knowledge based on retention as that doesn’t indicate how fast that estimated knowledge decays.
Stability isn’t any more “stable” than interval length though. It can change any time the memory state for a card is recalculated – which, for instance, happens every time you optimize, regardless of whether you reschedule (which might change the interval). I would think (outside of studying) Stability will change more often than interval for a typical user.
I would think you would value cards varying between young and mature then. When they shift, you are getting adjusted to more accurate information.
You can look at average stability. That’s least arbitrary.
I have 2 decks with different target retentions. It’s nice to be able to glance at each and see how much of each deck I remember well in a way which is consistent between the two decks (which I’m defining as approximately what percentage I won’t have forgotten if I don’t study for a while).
With the different target retentions the current interval based mature count becomes not useful for this. A stability based mature count would give a reasonable estimate at the number I care about.