Between those sites, you'll see breakdowns by gender, ballot delivery method, age, party affiliation, and a bunch of other variables.
A unique benefit of a nearly-completely distributed system like the US's is being able to look at, say, NV, GA, and PA. Their mail-in ballot composition relative to in-person is similar even though they're completely different processes and people.
You can download 446mb of clearly bad data.
This generally occurs over a week or two after the initial certification, on roughly the same schedule as the recount. We've seen this a few times already for early results and will see it more in the next few days. https://www.dos.pa.gov/VotingElections/OtherServicesEvents/V... and https://www.dos.pa.gov/VotingElections/OtherServicesEvents/V... have a bit more on this, though there's probably another PDF I can't readily find which defines the timeline to dispute/contest certain ballots.
In practice this means that both parties' national organizations have teams running voters' names through public databases right now, looking for dispute-worthy ballots/voters. I've never been a part of a national race so I can't describe the details of that, but my guess is they use the same data brokers as debt collectors, plus some data brokers that are specific to voter rolls. l2political.com is the best known of the political databases (details: https://guides.nyu.edu/l2political).
So, the ELI5 is: it's the next step in the process between receiving ballots and final tabulation.
(Source: Have volunteered as a ballot-counting observer in Seattle, volunteered for a local City Council candidate's campaign, and am leading a campaign to change Seattle's voting method. I'm nowhere near an expert but do know a lot more about it than most people.)
Why would they have been accepted to begin with?