The full machine comprehension of open domain text remains an unsolved yet central problem in Artificial Intelligence. We suggest a precise operational definition of the problem; we consider what the high level desiderata for such a system might be, and how far along we are to achieving them; and we offer seven signposts that seem worth bearing in mind, for those attempting new approaches to this task.