As the previous commentor said, this isn't point in time restore or a change in our processes. It is a change to our database and storage engine to mark regions of log-structured storage as if they did not occur. That's why we can complete a Backtrack in a few seconds - we're just moving some pointers around and restarting the database.