Wonky normals on some points using "brep closest point"

All these points were originally created by the “divide length” node, but what’s previewing now is from the “brep closest point”. The red circles are wonky normals on the same surface, and the green circles are ok normals:

image

I understand that a point on an edge might have an “undefined” normal, but why, on the same surface, would such big inconsistencies form? “Brep closest point” should put it at the closest point that still has a surface normal, shouldn’t it?