In SelDupAll_ignores_one_of_3_points.3dm (35.7 KB), 2 points are ~9e-11 apart and a 3rd point is ~1e-6 from the other 2. _SelDupAll selects the 3rd point along with only 1 of the other 2.
Okay? Floating-point math with tiny distances is fuzzy and weird and annoying, that’s computers for you. What is the actual practical problem you are trying to solve here? Chasing these exact 3 points down isn’t going to fix anything.
I check the quality of a brep using _SelDupAll to search for large distances between the endpoints of edges. After realizing that a pair of points was missed, I decided to report it. I added the 3rd point to emphasize the bug.