I would think that would show up with a summary discrepancy of the count of the ballots. First you’d match overall ballot counts and drill into the specific ballots when disagreement is found. It should be a straightforward merge operation on the ballot ID where null values reveal a missing datapoint for a given row in the ballot.