Some context from the repo readme, since it is a pretty click-baity title that's taken out of context.
BallotProof takes in input images of the front and back sides of your ballot and uses [client-side] image analysis to specifically determine which errors can prevent your critical vote from being counted.
ALL BALLOT IMAGES ARE AUTOGENERATED BY A COMPUTER FROM A SINGULAR SAMPLE BALLOT. THESE BALLOTS DO NOT EXIST PHYSICALLY AND ARE NOT INTENDED TO BE SUBMITTED AT A POLLING LOCATION OR BE SENT IN THE MAIL.
The generation script (generate.py) enables the generation of semi-randomized ballots that fit certain satisfiability criteria. We use these sample ballots as tests for model functionality.