SFoyil

Only criticisim is the sensitivity of your assumptions. If you assume 90% L1 continue, the figure jumps up to nearly 50%. If you assumed 75% continue (maybe more realistic?), the dropout goes down to 20%.

As you note, some of the "dropouts" are taking their 2nd or 3rd try, so the persistance of 50-80% seems reasonable as a cohort.

I think you ignore the "special cases" since there are both positive and negative count reasonings that cancel in a steady state.