Dynasty wrote:

Each is decomposed into two separate triangles. There are two possible different ways you can partition the face into two triangles. For each diagonal of the square, take the absolute difference in lighting levels between the two vertices on the diagonal. You should partition the face so that the set of diagonals with the *lowest absolute difference* in lighting form the hypotenuse of the triangles.

Very interesting! But it's not clear to me if this really fixes the problem or simply minimizes the extent to which it occurs (which is also nice of course)?

And so are you doing this during surface extraction or afterwards? I would imagine that during extraction would be the easiest time to switch the order of the triangles but you'd have had to dive a little deeper into PolyVox.