So I'm a little confused by the election results numbers coming out of San Francisco right now. The first-choice results on sfelections.sfgov.org/novemb… don't match the short RCV tables at sfelections.sfgov.org/novemb… and neither matches my calculations at dbaron.org/sf-elections-rcv/ . 1/2
1
The errors are pretty small -- but I'd have expected these things to match. I suppose the least bad explanation is that the reports were produced at slightly different times during a count that was adding ballots to the dataset. We'll see if the later results match... 2/2
1
OK, I figured out why my numbers didn't match the city's RCV results; I needed to discard marks that had the IsAmbiguous field set to true. So at least my RCV tables match the city's tables now.
1
2
Is this because SF moved from “complete the arrow” to “fill in the circle” ballots?
1
Well, they moved to an entirely new voting system that produces output in a different format that I needed to implement support for. I had a test dataset last week, but I didn't have the corresponding tables from the city to check my output against, so I needed to do that live.
1
1
Also, the amount of data that SF will be publishing for this election is going to be ridiculously large. I believe it will include images of every ballot.
1
Holy crap! Can’t wait for the, um, outliers to be found and shared!
1
Replying to @haroldliss
I already went through a bunch of the overvotes (multiple votes for the same rank) to debug the problem. A few people always vote in interesting ways. A bunch of fun things like "rank three candidates #5 and three candidates #6".

Nov 6, 2019 · 6:53 AM UTC

1
Replying to @davidbaron @haroldliss
The rule for counting ballots with stuff like that is that once you reach a rank with multiple candidates, the ballot doesn't count. So a ballot that says Alice #1, Bob #2, Christine #2, David #3 will count towards Alice until she's eliminated, and after that will not count.