[elmomle]

I disagree. If we're considering the debug/issue fix approach (which I quite like, because it displays code literacy and clarity of explanation in a context like that one would find in a typical workday), if I use the same task/tasks for each candidate, I'm implicitly (and tremendously) biasing myself toward a universe shaped by the tasks I chose. What I'm really looking for is a more universally distinguishably trait of lucidity and perspicuity (coupled with sufficient technical skill, which, again, bias and variance from your tests' structure is always an issue in accurately scoring), in any case.