The paper doesn't cite specific commits used. It's possible that any of the commits in stable are actually good commits and not part of the experiment. I support the ban/revert, I'm just pointing out there's a 3rd option you didn't touch on.