Unless you've already diagnosed and fixed the bug yourself, you really have no objective idea of how minor the bug is. And therefore I can't see how you could m

Unless you've already diagnosed and fixed the bug yourself, you really have no objective idea of how minor the bug is. And therefore I can't see how you could meaningfully judge candidates. If you're using a fresh bug with each candidate, that also decreases objectivity because you're not comparing apples to apples. Additionally, if the candidate gets totally mired you don't have any way to give them a nudge, because you don't know what to do either. Unless this is just a test of pair programming -- and without expecting to reach resolution of the bug -- it doesn't really sound useful to me.