Pretty simple math really. Firstly I took inspiration from this formula from SM-17.
So I started with this as my base formula (excuse me for my handwriting, I figured it would be easier for me to just write it all out).
But then this would give me the shape of a graph that I would not associate with what I would imagine to be the relationship between set number of intervals and retention on the next day.
However, this could be easily rectified if I just switch up the signs.
Here you can start to see the X axis could be my number of intervals and the y axis be my retention rate. It becomes asymptotic towards the end.
So here is my new formula.
So I need a constant which tries to depict my memory to work with. This cannot be achieved without actually having some data to begin with. So the more tries there is, the better. (kind of like FSRS needing at least 1000 reviews).
So I let R for example be 98% retention rate (which is absurdly high) and a set number of learning intervals be 4. The rest is just basic highschool level math.
I ended this as my memory constant. By repeating this multiple times and taking the average, it should become more accurate.I don’t know I could make this be more dynamic though.
Now I can go again, but this time with a constant being already present, I cannot now set my desired retention rate which then, by solving for the intervals set, would give me the number of learning steps that should be put into Anki.
I found this equation to be surprisingly accurate in depicting my next-day-retention. It doesn’t however tell me how I should space my steps or if they are efficient at all.
It holds up well to other simple memory curves that I made
Maybe all of this is just anecdotal, but maybe the developers could use this as an idea to begin with and start incorporating short-term-memory or learning intervals into FSRS as well.
Maybe @L.M.Sherlock, @Danika_Dakika @dae could come and have a look at this.