I’m just using the default normals of the mesh. The tangent slot is for skewing the anisotropic part of the material. I’ve checked the mesh distance fields and tested it on other geometry as well, no dice. I’ve checked a ton of lumen things like the cards and such, still no dice.
My guess is you’re right and that there’s some approximating going on when some value is below some threshold. Looks like it drops to a simpler algorithm or something.
Here’s a screenshot of that same material on a cube in the level. Looks like the math involves the dot product of the light source, which makes sense. The top is still working if you look carefully:
EDIT:
I guess it does help to have some kind of normal map to go along with it. Here’s what happens when I throw in a fake normal map to it (just reusing the anisotropic map +1 in the blue channel). It still doesn’t look like the back face there is doing things 100% correctly, but it at least masks some of the issue and doesn’t looks glaringly flat:
