We propose a simple, tongue-in-cheek behavioral capacity model of contestants rooted in intrinsic motivation and feedback loops. Based on that we theorize what organizers can hope to get out of a contest, for what kinds of problems the contest approach works. It is shown via real life examples how perfectly rational behavior of contestants may lead to undesirable outcomes, how the rules, evaluation metric, available data and information may conspire to reduce utility and what can be done to address those issues. We also touch upon the limitations of current classes of contest problems: what makes them all too similar, and have a glimpse at new frontiers.