> How can we know for sure the binaries running on the voting machines [..]

As far as elections are concerned, give me paper ballots every day, and make sure you count them by hand with plenty of Mk I human observers present.

In Brazil there is black box testing. Party officials can choose a sample of a few ballots per polling location on election day, where a simulated election is run, with all inputs recorded on video. The machines have to tally the right votes for the input given at the end of day. These are machines that would be used for voting, and the party officials are parties unrelated to (and in a way, antagonistic with) the voting authority.

I think it's a pretty solid scheme.

Yeah this is why you shouldn't use voting machines. Paper voting is already great. Whoever is trying to sell you a voting machine does not have your best interests in mind.

Very few people have any change of understanding machine voting systems. With paper voting we get much better transparency of the voting procedures. Any form of machine voting is terrible for transparency and democracy compared to just plain old paper voting.

Could you not compile them on the voting machines itself? But yes, there is always going to be some level of trust involved, and the bar for manipulation seems to be lower than re: manual counting.

That's not really a "hole" - rather it's the idea not covering every possible form of corruption.

There's reproducible builds project for that. (Except too few people will know how to actually verify it.)