Or maybe it was the content of his posts flagging the right ML features. At the scale of any large entity, you're going to get things that fall through the cracks because of human error or machine error. Unless the mistakes are happening all the time, how they're handled is more interesting than that they happened.