I don't quite understand why the expect that there would be ~6x the number detected, though, assuming that the ~8500 cases detected is accurate. It would be very (and probably statistically naive) if the minimum total cases was simply because they have only ~1/6 of the total number of state pairings.

I think the other major concern I have, other than the methodology, are the definitions - I still don't know whether 8500 represents 8500 people who voted twice (17000 total votes cast), or 4250 people who voted twice, or something in between, or some thing completely different. Perhaps I missed this.