For earlier cycles, I can do what would've been most surprising at the start of the cycle, but many of the wackiest results of the 2010s that no one would've predicted a year earlier (like MO-Sen 2012, or either the AL or MA special elections) were accurately forecasted by polling.
Missouri 2012 actually wasn't forecasted by polling, which is really weird. "Legitimate rape" appeared to only switch it from a narrow Akin lead to a narrow McCaskill lead. With the exception of a single +15 outlier, the polls generally pointed to a likely result of about McCaskill +6 (rather than the McCaskill +16 that actually happened).