Triangle winding

The index buffer of a mesh is a plain triangle list (Model libraries). Nothing in the file says which side of a triangle is the front, so the side is found by comparing the order of its three indices with the stored normals.

For each stored triangle the geometric normal (b − a) × (c − a) was computed from the positions in index order, and its sign was compared with the average of the three stored normals. Over all rigid meshes of the disc: Observed

Result	Triangles
the geometric normal opposes the stored normals	5,384,944
the geometric normal agrees with the stored normals	1,469
ambiguous (the stored normals do not decide)	23,293
degenerate or with a zero average normal	7,078
stored triangles in all	5,416,784

So 5,384,944 of the 5,386,413 decisive triangles, 99.97%, are wound clockwise relative to the stored normals: the indices run in the direction that makes the geometric normal point away from the normal the vertices store. Proven · all stored triangles of the rigid meshes The same counts result whether the mesh is read in its own frame or under a mirroring frame. Observed The 1,469 agreeing triangles are not explained. Unknown

A reader that wants counter-clockwise front faces swaps two indices of every triangle, and does so once more for a placement whose matrix has a negative determinant (Placements): the mirror turns a clockwise triangle into a counter-clockwise one. Inferred