"Got lucky once" is one of the craziest descriptions of multiple models across multiple domains that demonstrate consistent high rates of calibration across time. Yes his initial success in one election was likely at least slightly anomalous, and that's the thing that made him famous initially. But to pretend that he has not continued to have continued as a predictive modeller is just wild.

While he is no longer with "fivethirtyeight", this is from when he was:

https://projects.fivethirtyeight.com/checking-our-work/