How polls handle people who have already voted is interesting. In theory, one could argue along these lines: people who say they've already voted should get weighted at 100%, while people who are merely "likely voters" should be 90% or whatever since some won't wind up voting.

In practice, few pollsters use probabilistic turnout weights. And even if they did, you might not want to weight "already voted" voters at 100%. There's some rate of ballot spoilage in mail voting (also in person, though less). And people could lie about already having voted.

Still, this presents challenges for likely voter models with early/mail voting being highly D-leaning this year. I can imagine various strategies that pollsters would adopt that could lead them to either underestimate or overestimate Democratic margins.