Okay this part is where i kind of fall on my face. I don't know how the vertex coordinate system works precisely but you can see what's happening in the following patch.

Each cell has 2 planes and it's dim is 20 20 so I'm assuming that the vertex coordinates are as follows:

Plane 0, cell 0 19 val 0

Plane 1 cell 0 19 val 0

Plane 0 cell 0 0 val 0

Plane 1 cell 0 0 val 1

Plane 0, cell 19 0 val 1

Plane 1 cell 19 0 val 0

Plane 0 cell 19 19 val 1

Plane 1 cell 19 19 val 1

Meaning from the bottom left clockwise you'll see (0,0;0,1;1,0;1,1) which are normalized corrdinates (but different than the standard jitter matrix which is usually represented as top-left, top-right, bottom-left, bottom-right.

So jit.matrix sets up a matrix with 2 planes, each cell represents an X, Y coordinate of dimensions 20 x 20. Jit.expr normalizes values across X on plane 0 (which represents the X in each cell) using (norm[0]) and then normalizes and subtracts the value from 1. down the Y plane on plane 1 (which represents the Y in each cell) using (1. -norm[1]).

But this is my educated guess based on what I'm seeing... I could be mistaken.