The book Thinking in Bets points out that predictions are given with a % of certainty. If I say X party will win at Y% and Z party wins, it doesn't mean the model is broken.
Given two major elections in a row where the results were essentially out of the error bars, it is perfectly reasonable to be dropping your confidence on pollsters very hard. Once was perhaps understandable, but twice is getting into "right .01% of the time" territory.
At a poker table you get a high number of events to check against but a presidential election is only once every 4 years.
I wonder how accurate the local elections are vs the national? If neither is accurate, then that's a strong case for your point of view and a change in methodology.
I wonder how many people still think Nate Silver has a credible perspective: he's spun the above strategy into a job in unscientific punditry.
Correlated errors are tough!
The model certainly ought to handle correlations, but the evaluation should take them into account too.
You can't treat the election as fifty independent replicates: a miss in both Minnesota and Wisconsin is clearly worse than one error, but it's also not as bad as (say) getting Wisconsin and Rhode Island wrong, where it's more likely that two separate errors occurred.