Yeah, I think the complexity would come from the ‘decay’ varying from person to person, so you’d end up with a results matrix of accuracy vs accuracy decay.
I was imagining a person’s overall accuracy declining, but I suppose if you assume that most of that decline comes from higher SRS-level reviews then you could model it on a by-item basis. Even if there are other factors, such as “more items = more potential for confusion between items”, I guess you could just account for that by varying the modifier you use I always sucked at this kind of maths
I was thinking about it because if I’d looked at this when I was a wee level 10 I would have underestimated how many more reviews I had ahead of me - my accuracy has declined since then.
I don’t know if this will affect the outcome, but in reality the success rate is not independent of the vocab/kanji.
Some will have near 100% success rate, others much lower, averaging out at the overall success rate.
Also, assuming that the SRS method actually works, these rates will likely improve over time (at different speeds ofc…).
Also, some words might be leeches that have success rates too low to realistically ever burn, effectively making “time to burn everything” infinite for some people
This would affect the outcome, at least of the first simulation, since it would change the distribution of failed reviews. This is not something which I’m looking to take into consideration though.
It would be nice to see some level 60s weigh in on the accuracy of the results. How do the numbers look to for you?
If you want to make the math even more complicated you could try including the times of each leveling to work out how many more months you will need before burning everything, depending on accuracy.
Are we nearing chaos levels of variables yet? Quite possibly!
I always sucked at this kind of maths