Can We Be Wrong? The Problem of Textual Evidence in a Time of Data