> If the numbers reported by the machine are nonsensical

The problem is that this is not a sufficient criteria for detection. Presumably a hacked machine would spit out "sensical" counts that are biased to one side, not 99% for one party or something like that. How would you detect that by just looking at a simple sum?

I can't really see how to verify the count without having everyone check that the machine printed what they thought, and having multiple people perform independent counts that must match.

You could have people count later and check that the machine got it right, but then consider the problems that would cause if the media was reporting bad numbers before the official count is finished.

The point of that remark was to point at the ballots being the actual record of the vote rather than the result provided by the counting machine.

It is certainly the case that trust in such machines could be misplaced.

> It is certainly the case that trust in such machines could be misplaced.

Right but if that's the case, you're going to have to count them by hand at some point. So what do the machines bring to the table, if their whole purpose is to avoid that?

You usually only need to hand count a small number of randomly selected votes to verify the result statistically. The exception being very close results where a handful of votes makes the difference.

That is a good point, as long as "randomly selected" can be assured.

So, just make the machines nationwide present "sensible" but false results?

Well, in addition to (just) hacking all the machines, you'd have to tamper with a sufficient number of exit polls to limit recounts (and the waves of subsequent recounts that would follow widespread evidence of tampering).