Now I understand it - the maximum spacing between the edges of original surfaces equals double the fillet radius. The Extend option of the FilletSrf command does not solve this problem - maybe it should. The FilletSrf documentation mentions the rolling ball analogy, but it does not say that the FilletSrf may fail because the gap between the original surfaces is too wide. If the gap is too wide, Rhino displays cryptic message: FilletSrf failed to create fillets.
(I have spent lots of time withKelvin Cheng fixing bugs in Rhino documentation, but there is still about one dozen errors and about one hundred omissions.)
Well, no, that’s not it either. There is no location on the edges of your surfaces where they are anywhere near .22 units apart. You’re onto something, though, with the rolling ball idea. That rolling ball is what determines (so to speak) where the rails will fall, and if they fall off of the trimmed face, you get failure.
extend option means something else: this is not extending the original surfaces, but only the fillet surface as far as it can reach on both surfaces. I think what you expect though is that Rhino would extend surfaces just like it extends curves with the Fillet command?
Yes, both surfaces should be extended (ExtendSrf) by the same distance so that the ball can roll on them.
I’m not expecting that to happen soon if ever.
Btw: One way to check if a FilletSrf will fail is by offsetting both surfaces (at the side of the fillet) and see if they fully intersect.