so the edge that minimizes the quadric error is also the edge with the optimal point that is closest to triangles it covers from the original mesh.

Here are we assuming that the normal of each plane is unique? If not, we should have infinitely many quadric error matrices for each triangle plane. But if it is unique, how do we determine which normal to pick?

@azul since we're normalizing the normal, I assume there's only one/"unique"?