I once said to one of my students (who was convinced that their code did not have a particular bug they could not reproduce):

Start from the assumption that the reported problem does exist