Interesting video, but I'm not sure about the methodology here. Adding polling error to polling that may have already been error-corrected will double count the errors, and we don't know how pollsters methodologies have changed in aggregate. We don't have enough information.
If you used this exact same methodology in 2020 (literally beat for beat exactly the same way he did it in the video), it would have predicted Trump to win because of how much the polls underestimated him in 2016. But last time the polls were only out on average by 2 points, so you'd have assumed an average polling error almost three times the size of the actual error we saw in the election.
Don't get me wrong, we're all guessing here, but this isn't any more sophisticated than just guessing with extra steps. You can't use historic error to predict future error because if we could then we wouldn't make errors in the first place, it's a bit like a form of Gambler's Fallacy.