I think you understand now what I have written. Let's forget about the rest here.
I just saw that yes -- I have messed up the test with the original tolerance value too. However setting the tolerance to 0.0001 still clears up the issue and practically that number is absolutely okay, for sure it's better than chamfering the edges.Edit: I was wrong changing the tolerance to 2.09e-7 did not lead to a clean BOP check, although it did give different results, turns out the first time I was just running the check on the single edge since I hadn't reselected the body I suspect the same happened to OP here so it might have happened before too.
The final floating point accuracy depends on a number of things, it's not a fixed number - so whatever OCC is calculating inside might have an effect on this.
I attached the testsample as a playground (I just created it a few minutes ago - and no I did not have that before either).