>This whole watermarked thing is a LARP.
The version of the theory they could be true involves yellow-dot codes. They had to print and package the correct version of each local ballot for each addressee. How hard is it to record individual hidden marks and addresses they were sent to? It's not impossible that each mail-in ballot was not as anonymous as we usually expect them to be.
My belief in this is only about %15. Validating more than a random sampling by pulling ballots and contacting the voter is impractical. But it's the one version of the theory that is possible.