For what its worth, this appears to be some limitation of the TriangulateClosedPolyline method:
181018_problemfillmeshgaps_TroubleShootAHD_'00.gh (8.9 KB)
Perhaps something for the McNeelies to have a look at? Or maybe it just wasn’t designed to handle cases such as this? Either way, you probably want to weld your edges and ensure that your normals are unified and such, before running either of the suggested algorithms:
