Modern machines differ from old ones is a very significant way: they give you a paper ballot that you - the voter - can verify before submitting.
Assuming this is true, it would be simple to modify the printing software such that the QR code vote is different from the human readable vote.
A recount, by running the paper ballots through the machine again, would come up with the same result as the first time.
The only way to catch it would be to hand-count the paper ballots and compare the results to the machine counted votes.
Really though, the QR code should be eliminated, so that the voter is able to personally verify what the machine will actually read.
Probably as many as review their grocery store receipts, I'd guess. Which is to say, practically none.
But regardless of the above -- the design change to include printed ballots was a big improvement in security and voter confidence in the election.
The QR code could mis-match the printed ballot on say 5% of the ballots and nobody would notice a thing.
Need a recount? Ok, just run em through again, and they'll read the same (wrong) votes again, no problem.
Only a hand count of the paper ballots would catch the mis-match.
Did that actually happen? I don't know. But if the machines are truly reading a QR code, that seems a glaring flaw regardless.
Because according to what I saw presented in hearings, the reading machines actually read the QR code made by ballot-marking-device, not the human readable portion of the ballot.
If true, that seems an obvious design flaw (to be nice) right there.