[jpap]

Agreed; I do this all the time as well. I don't think it as "guessing" per se, I think of it as forming a hypothesis based on the intersection of my existing knowledge of the problem and any new observations [1], then testing it to figure out if the hypothesis was correct. If proven incorrect, the process helps me reform into a new hypothesis and the cycle continues until the problem area becomes clear enough to solve outright.

[1] By reading more of the code, RTFM, observing logs, tracing with a debugger, varying inputs, etc.