To reply to Kenobi's thing before I actually get some sleep: I'm not talking about the percentage chance thing or the margin which I've always found laughable enough to stick into the mode where it swings around all over the place because it amuses me - although notably they were always pointing in the right direction even if the impact was exaggerated while you, eh, called the election for the wrong man. However the other bit of the model: the bit about what votes were remaining to be counted and what already had been counted were perfectly right in terms of general trend if not magnitude: and paying attention to that is what I think made those of us watching elsewhere feel pretty sure pretty early on that things were looking like a very close result but with Jones in front which is what we got. The wild swings seems to have been because of turnout differentials more than anything else; at least that's what I thought at the time. Again usually they do this with races where they have precinct level data which is a million times better but they did the best they could with what they had.
Your model caused you to make the wrong decision, which means that either its core principles are wrong; the assumptions that you based it on are wrong or that this election was incredibly odd and unusual. I don't think that its the latter: indeed a fair few of us thought that it felt oddly similar the to Virginia Governors election: in terms of the fact that it seemed that Joneswas leading most of the night despite being behind until 90% of the votes had been counted. So therefore it must be one of the first two: and after getting a good nights sleep you probably ought to have a close look at it.
Incidentally the only way that Moore can get a recount is if it closes within 0.5%: so he basically needs to find something like 14,000 votes with zero election day votes to be counted. The provision that allows candidates to pay for a recount only seems to apply to State offices, and not Federal ones.
By the way; this was an election worth staying up until 6am for!
I happened to check the counties when Jones caught up (at around 86% of precincts reporting).
Baldwin 94%
Franklin 42%
Jefferson 73%
Madison 90%
Mobile 57%
Montgomery 49%
Shelby 63%
All others 100%
So you had 6 of the 7 largest counties still out (Tuscaloosa is 6th and was done). The Top 7 are Jefferson, Mobile, Madison, Montgomery, Shelby, Tuscaloosa, and Baldwin.
Franklin was the only outlier, and when I checked again a few minutes later it was 100%, while Baldwin, Jefferson, Montgomery, and Shelby were unchanged.
Madison had bumped from 90 to 95 and Mobile from 57 to 70.
If we project the results for the late votes based on the total votes (e.g. since Jefferson was at 73%, we take 27% of the final results and attribute them to the candidates), then Jones won the last to be reported votes by a 63.4% to 34.9% vote.
If we project after the changes for Franklin, Madison, and Mobile then it was 64.1% to 34.1%.
There was nothing wrong with Ben Kenobi's model other than sample bias. His 86% sample was not representative.
There was quite possibly a selection bias. Smaller counties are likely to have an older electorate because students go off to college elsewhere and don't come back if there are no jobs for their skills. And those who don't go to college may also leave if they can't find work. Older people can vote during the day. Also people can get home to vote after work in 15 minutes. If you come home after 45 minutes in traffic, you may need to rest before going out to vote. In smaller towns, someone may be able to vote at lunch or a slightly long coffee break. So there would be less chance of lines at closing time. Voting can be wrapped up at 7 pm. Since votes are counted on scanners, election judges might have to record a count at the polling place, but the reported count will take place at the county level when the cartridge or whatever voting device is stuck into the totalizing machine. There is a drive across town, and then maybe a wait in line as dozens of precincts wait in line. In a smaller county, this can be done much quicker.
And all of this assumes that there was not differential turnout. Each irregular black voter is worth 0.9 votes for Jones. All have heard of Trump. Each irregular white voter is worth 0.4 votes for Moore. And few will have heard of Schumer. The payback is much less for the white voter, and the trigger is higher.