I've never used a sandwich construction before, and I was excited thinking I was finally going to for this challenge... but a more straightforward version ended up shorter. :( 13 bytes vs 11 bytes.
I've also got 11 bytes by porting everybody who's doing int-division to calculate the repcount directly: lᵐ÷+₁;?tᵗj₍ or lᵐ÷+₁N&tj↙N (both with swapped inputs)