Did anyone try to just blend pixels together as if they were round circles slightly bigger (or smaller) than a pixel box? I suspect that’s what CRTs actually did, because no point on CRT was a square (you can’t get a square electron beam).

Feels like these scaling methods solve a wrong problem, cause I’ve never seen one that would look even remotely like CRT.

What subj method does is basically adding scanline gaps to the blur, but only to the point of vertical isolation. Maybe the answer is to do something like this, assuming 8x8 grid and 0-9 intensity:

  2|33333333|2
   +————————+
  4|66666666|4
  7|88888888|7
  6|99999999|6
  6|99999999|6
  …
  (mirrored at the bottom)

  R G/B\R G/…
   B/R G\B/…

Sizing this “circle” would simulate the “focus” parameter of a CRT, which wasn’t available to a user.