Fix bug in discrepancy calculation for ES&S overvotes #2162
Conversation
Treat overvotes as 1s rather than 0s when calculating the numerical discrepancy
| deltas = { | ||
| choice.id: int(reported[choice.id]) - int(audited[choice.id]) | ||
| for choice in contest.choices | ||
| } |
There was a problem hiding this comment.
6ddd24c to
bef3648
Compare
| return None | ||
| else: | ||
| deltas = { | ||
| choice.id: 1 - int(audited[choice.id]) for choice in contest.choices |
There was a problem hiding this comment.
This is a little weird, though it's closer to what the math does. The math treats an overvote as a vote for the winner and a vote for the loser, whereas this logic treats an overvote as a vote for every candidate.
arsalansufi
left a comment
There was a problem hiding this comment.
I know we're opening up a broader discussion about overvote handling, but what you've presented in this PR does seem sound and at least consistent with the current math
| # delta. Otherwise, return discrepancies as usual, but substituting in | ||
| # overvotes/undervotes. | ||
| if has_overvote: | ||
| if audited_votes > 1: |
There was a problem hiding this comment.
Hm I know this is just pulled from the audit math but do we need to account for the case that a contest allows for more than 1 vote?
There was a problem hiding this comment.
ES&S CVRs only suport vote-for-1 so far: https://github.com/votingworks/arlo/blob/jonah/fix-ess-cvr-error/server/api/cvrs.py#L851-L853
We might want to add an assertion here and in the audit math about that assumption in case it changes
| choice.id: 1 - int(audited[choice.id]) for choice in contest.choices | ||
| } | ||
| elif has_undervote: | ||
| if audited_votes < 1: |
There was a problem hiding this comment.
I'm assuming undervote here refers to a completely undervoted, i.e., blank contest. And not a vote for N where someone didn't vote for 0 but just some number less than N
|
Pausing for now per #2100 (comment) |
Alternate idea to fix #2100 that mimics the logic from
supersimple.discrepancy