Let's say you're trying to find the cause of a bug between 3.11.0 and 3.12.0. The bug is really difficult to catch, because it may or may not manifest itself over several days of regular workload uptime.
You mark 3.11.0 good. You mark 3.12.0 bad. Start bisect.
Then, running on the bisect patch, the kernel fails in some other significant way that makes catching the original bug maybe impossible. For example with some constant drm graphics corruption it may be impossible to replicate the several days of uptime needed to try to catch the original bug.
So what do you do? Can't really mark this patch "good" right, so the logical option would only be "bad".