Testing What You Can’t See: Risk Blindness in Coverage Models