Programming students who encounter performance bugs often struggle to fix them and sometimes resort to less-than-ideal debugging methods such as random guess-and-check or asking GenAI to give them the answer. This three-phase dissertation aims to help students understand, find, and fix performance bugs more effectively.

The first phase studied the problem, developing a taxonomy of common performance bugs consisting of 3 categories and 12 sub-categories through qualitative analysis of 250 slow student submissions. The second phase devised an intervention, a novice-friendly Python profiler called Hypothesis Profiler (HyProf), deployed it in a 400-student Python course, and evaluated it through web logs, office hours observations, and surveys. The third and current phase is evolving and evaluating the intervention, by using machine-learning techniques to automatically create function-specific performance bug labels in HyProf reports for slow submissions, similar to how exceptions label the type and source of runtime errors.