This is also what bothers me. The example before the one you mentioned looked the best to me. And it seemed to reliably detect numbers and operators as well, so I don't see why this approach should not be able to resolve the equation like the last example did.
Writing in monospace is not only unnatural, but the approach shown even makes it one step worse: you need to write in the cells. You're completely giving up the flexibility of handwriting, you need to know your equation before you write it down, otherwise you're risking running out of "cells" on the left or right.
Yes, the grid paper is a stepping stone for the line paper. However, it is a tradeoff. With grid cells you can tell what symbol is in it. With symbols on a line, you have do guess a lot. Illustrative example: there is a unicode symbol for "colon equal" – "≔" – if you write it in a cell you can be sure you got it right. If you write it on a line you could mean two distinct symbols – ":=".
The problem you mentioned – that you have to plan ahead – is not true for grid cells. There is a part in the GRAIL system demo where it is shown that you can go quite far with this approach. [0]
But as you mentioned, flexibility of handwriting is in its non-modal interactions. I am currently working on a line-based editing, but it's a bit tricky.
Writing in monospace is not only unnatural, but the approach shown even makes it one step worse: you need to write in the cells. You're completely giving up the flexibility of handwriting, you need to know your equation before you write it down, otherwise you're risking running out of "cells" on the left or right.