There's a chance that this is a statistical anomaly, like "polling" your bag of M&Ms, pulling out 5 pieces and getting all blue ones. This one had that usual "95% likely to be within 3.4% based in 808 likely voters" disclaimer. Was this not merely part of the 5%, but at the 0.001% level? Or, more likely in my opinion because it was the very last poll before the election, was this tweaked to get a result in Iowa 16% (9.4 standard deviations) off from the actual vote to try to affect the election.
It is hard to create this big of error without leaving a bunch of fingerprints like editing call lists to eliminate Republicans or just outright ignoring call results and making soup numbers. Discovery should be interesting.