Show me one time when exit polls are 100% right.
All polls are wrong to some extent. Add in a flawed model (and in this case, results taken only partway through voting!), and they'll be even more wrong. Have them all done by the same organization, and, if it's wrong in one place, it's probably going to be off in another.
'A million to one?' Let's see the math on that one.
In the precinct where I was a watcher only 1 Republican voted on election day. They all voted early and Republican turnout was huge.