Any historical track record of how much late votes were more Democratic precinct by precinct, or whatever, than earlier counted votes? It would need to be a big discrepancy in favor of the Dems to make the math work for Begich. The late votes were more Dem last time than earlier counted votes, although not enough to close the gap for Begich if replicated by a couple of thousand votes, but that is raw numbers, not corrected for whether or not late counted votes were disproportionately from precincts favoring one party or the other. Intuitively one would think late votes are more more Dem precincts, in which event Begich is fried per the above number crunching projection, but sometimes intuition about a state you know next to nothing about psephologically, is a fail.
Begich turned a 3,000 vote election night deficit into a 4,000 vote victory in 2008. But there were 90,000-100,000 outstanding post-election day votes in 2008. This time, there are only 50,000 at most - and that's probably optimistic.
In 2008, Begich won absentees by 6.7 points and the early vote by 21.4 points. So far, in this election, Begich is LOSING absentees by 6.9 points and only winning the early vote by 8.7 points. Granted, there are more absentee and early votes to count, and the current margin might not be representative of the final margin, but I doubt the margins will swing by 15 points. There also are an unknown number of questioned votes. In 2008, Begich only won them by 3.3 points - slightly better than his overall margin, but not terribly so.
Baring some huge miracle, Begich is toast.