We often point to the relative increase in the amount and sophistication of evaluations of visualization systems versus the earliest days of the field as evidence that we are maturing as a field. I am not so convinced. In particular, I feel that evaluations of visualizations, as they are ordinarily performed in the field or asked for by reviewers, fail to tell us very much that is useful or transferable about visualization systems, regardless of the statistical rigor or ecological validity of the evaluation. Through a series of thought experiments, I show how our current conceptions of visualization evaluations can be incomplete, capricious, or useless for the goal of furthering the field, more in line with the "heroic age" of medical science than the rigorous evidence-based field we might aspire to be. I conclude by suggesting that our models for designing evaluations, and our priorities as a field, should be revisited.