@phkahler the disk of the machine I usually use to work on SolveSpace died... :-(

Thinking "on paper" - how would the caller choose the dir parameter?
Apart from that I need to "parse" more carefully through the dot an cross products and make some drawings - but at first glance in seems that the function may return a positive number if the surface "curves towards" the dir vector. I think there are some corner cases...

Thinking "on paper" - how would the caller choose the dir parameter?

@ruevs My thinking is that many NURBS issues are a result of surfaces being classified as "coincident" at an edge because they have the same or opposite normal. In those cases we need to disambiguate be taking curvature into account. One area to call this function would be in SSurface::EdgeNormalsWithinSurface
near the end. That commented out change should work, but the calling function will need to consider curvature so maybe it should be calculated there and returned - I'm not sure where exactly to call it, just how it needs to be used. It seems to be what's needed for #1291 for example, where the fillet fails if its tangent to the cube face but not otherwise.