Given coverage data, how would you build a tool to decide which untested parts of a codebase most need tests? The best I can think of is using profile data to ensure hotspots are tested. Not sure if it's ideal though: top level code and well-exercised logic would be highlighted.