I don't think this is a good description of Silver's model. It doesn't discount disruptive events at all. It implicitly takes them into account.
Silver knows that people's opinions can be changed by events that happen before the election. This is why polls taken early on are less informative about the final result than polls taken closer to the day. Based on historical records, Silver knows how the accuracy of polls depends on the date they are taken, and he weights them accordingly. This process automatically models the possibility that an event could suddenly change people's opinions just before the election. It's taken into account in terms of the accuracy of polls.
Silver knows that people's opinions can be changed by events that happen before the election. This is why polls taken early on are less informative about the final result than polls taken closer to the day. Based on historical records, Silver knows how the accuracy of polls depends on the date they are taken, and he weights them accordingly. This process automatically models the possibility that an event could suddenly change people's opinions just before the election. It's taken into account in terms of the accuracy of polls.