Not just collusion, but collusion at scale. See my description of the Minnesota process. Voting and counting are done at the precinct level, and there are thousands of precincts. It might be possible to subvert a single precinct, in order to manipulate its outcome. But how many precincts would need to be subverted in order to affect the outcome in a statistically meaningful way?

Additionally, can the precinct be subverted in a manner that can withstand outside auditing by non-corrupt district/state-wide election officials? Keep in mind that if one party in the election has substantial reason to suspect the results were broadly rigged, they could demand a recount (even at their own cost, as happened with the Dayton/Emmer recount in MN), thus triggering all those downstream audit controls.