he starts trying to prove a direct relationship between the eventual x co-ordinate on the screen and its originating z co-ordinate in world space
All coordinates are in camera space.
There's a linear relationship of x,y,z because it's a triangle.
So there's one z coordinate for every pair of x,y.
A triangle is drawn top to bottom. In each scanline only x changes and y keeps constant.
So to say there's one z for every x (of a scanline).
Once (x,y) is 2d-projected, the linear relationship of (x,y) -> z breaks.
Instead, 1/z is linear across the projected triangle because it z gets the same transformation as x & y.