1. I have seen video demonstrations wherein one person with opportunistic unsupervised access to a voting machine can hack it to affect every issue on the ballot within five minutes. Given the security at many polling places in the US, this easily extrapolates to one person able to hack all machines at a given polling place prior to election day, and possibly repeat the effort at several other nearby polling places within counties using the same type of machine.
Other hacks target the central tabulators, and don't require touching the voting machines at all. Opportunistic timing to enjoy unsupervised access between the election and results certification is a bit more critical to these.
2. Exit polls have frequently been "adjusted" in the US to more closely match the actual elections results. In the UK, this is called the Shy Tory Factor[0] or Shy Labour Factor, and has existed since 1992. The US has had the Tom Bradley Effect[1] since 1982.
This is usually explained by presuming that the exit polling result is inaccurate. It is never seriously considered in public that vote fraud may be becoming more severe. Yet I saw a statistical analysis of actual voting results, differentiated by the model of voting machine, that showed a clear, obvious bias towards certain choices of specific issues. The poll adjustment factor can easily be explained by the hypothesis that people remain as honest as ever in exit polls, but the voting process itself has become less honest.
3. Recounts will not uncover fraud if the records necessary for the recount are subject to the same controls as the records used for the first official count. If the voting machine does not leave a non-electronic record that can be verified by the voter while they are still at the polling place, a recount will just be run on whatever records exist, which may be falsified.
4. Thanks to the winner-take-all elections in the US, and the availability of previous census and elections data, it is possible to identify specific counties as keystones, wherein efforts to sway the vote locally will have disproportionately effective results.
For instance, you examine the "battleground counties" in the "swing states", and hack the central tabulator in that county to give your favorite candidate a 3% advantage by switching an opposing vote 1.5% of the time. You don't tell your party you're doing it. You don't even tell your dog that you're doing it. You just do it. If your guy loses, well, you tried.
Multiply by dozens of technically competent yet unconnected and independent supporters, and you get a significant extent that does not require any team.
