I would agree with Jake's position that discrepancies don't always point to fraud.

My particular district had something like 1400+ voters cast ballots, and many of those ballot "styles" were multiple pages (front and back) because of the number of questions on the ballot as well as the english/spanish translations of each question.

So, roughly 2800 ballot "pages" were run through the machine (not counting the "provisional ballots" which are counted separately).

We also have to consider that more than a handful of voters screwed up their ballot (didn't fill out the bubble, marked areas that the scanner couldn't read, etc) and had to "spoil" one page and get a new copy.

Heck, some voters intentionally did not fill out anything on their ballots. There is even a procedure to count those blank ballots...

Each of these "issues" has a corrective action (and documentation) which helps prevent or at least explain any difference between the number of voters and the number of ballots cast.

Is all of this perfect? I doubt it. But does this constitute "fraud", or will this difference be enough to swing an election by 2 million votes? I suspect the answer is "no".