Test it a million times and the poll book fails: Ginny Badanes on Los Angeles' new voting system
When Los Angeles County used its new voting system for the first time in the March primary, some voters waited in long lines. The new ballot equipment was not the bottleneck. The lines formed before people ever reached it, at the electronic poll books, where workers check in each voter.
Ginny Badanes explained why the poll books mattered so much. The county had made two choices for voters' convenience: vote centers, where you can vote at any location in the county, and voting over several days rather than one. Both mean that to make sure each person votes only once, the record of who has voted has to sync between every vote center, continuously. From what she had read, and she was careful to say she had no inside knowledge beyond the county's own review of the primary, that is where the trouble was. When the poll books were not syncing or were slowing down, voters backed up before they could get to the machines.
You think that you've got it all nailed down because you have this great system that you've tested a million times, and then it's the poll book that causes the issue.
There are other places an election can fail, and she pointed to the next one: people. Elections run on volunteers, and this year the typical poll worker, often older and at higher risk from the virus, is staying home. There is a big push to recruit younger people to replace them, because even with far more voting by mail, people will still show up on election day and someone has to run the polls. You can plan for everything you think might happen, she said, and then a pandemic throws the plans in the air.
I raised a second-hand account I had read, and said it was second-hand. Poll workers trained on older equipment did what people do when a device misbehaves: turned it off and on again. A restart triggered a full resync of the database, and enough restarts at once loaded the network further, on what was probably one of the first times that network had carried this kind of traffic. Ginny's reading was that whatever the exact cause, this was less a technology problem than a process and training problem. Poll workers often work election after election and know the old system well. Los Angeles changed theirs radically: suddenly there were no paper poll books. A wave of brand new poll workers adds another variable, so training has to cover the process, not only the machine.
From an engineering point of view this is the failure everyone who ships systems learns eventually. You test the component you are proud of until it is perfect, and the outage comes from the seam: a network that has never carried production load, a synchronization step under real concurrency, and an operator habit nobody wrote into the runbook. The lesson is to test the whole path at full load with the people who will run it, and to plan for what a tired volunteer will do when the screen freezes.
California is mailing a ballot to every registered voter this fall, which should take some pressure off the vote centers. The seams will still be there, and that is where the next problems will come from.
From Lincoln Shorts, the interview series I hosted for Lincoln Network, recorded 1 September 2020.
