Sometimes I barely go around the debugging loop at all, once I have the problem reproduced. I find myself immediately forming a hypothesis, and testing it, and I was right on my first try, and now I understand the problem.

Maybe part of it is that, if I've managed to reproduce the problem, I've collapsed the solution space significantly, just by observing it in a controlled, known situation.

It might have something also to do with perception as well. Words on a screen are words on a screen, but if I had to interact with the system to get the error message to come up, I might perceive it differently than if the error message was shown to me by someone else.