Being an elections observer in King County, WA

My experience of being an elections observer at the absentee ballot satellite counting office at 3901 1st Ave South in Seattle.

I got a good look at the Accuvote counters, and observed that they had paper tape tallies like the ones at the precincts, so that counter tallies can be reconciled with the machine count. The people I talked to were all worker bees, and did not know when or how the reconciliation would take place. Batches of 20 or so ballots are fed to the machine, and if one doesn’t go through, the machine stops and it doesn’t get counted. Sometimes flattening the fold creases further or putting the ballot in from its opposite end will get it accepted. Ballots that won’t go through at all are held out for duplication. The hand count of the number of ballots counted is then reconciled with the printed tally, and if there is a disagreement, there is a hand count by a person selected for that task, in this case a young woman who had a lot of trouble getting ballots through her machine. If the second hand count agreed with the tape tally, the results were sent to the central processor (in the same room as the Accuvote counters, locked in a cage). If it did not, the results were not tabulated and the batch was recounted on another machine. Runners delivered boxes to be counted and checked on the agreement of the hand count with the tally strip.

I oversaw the ballot duplication process, in which ballots that could not be counted by machine were duplicated to be re-run later. The two people we observed had been through a training session, but were new to the process and could not answer my question as to whether the process had always been run that way or not.

They were able to tell me some things about the overall process, though. Signature comparisons were done at another facility near the airport. The envelopes are sorted by precinct, and then opened. Both the signature envelopes and the unlabelled secrecy envelopes are saved for later examination and counting, should that be necessary. At this point, ballots are examined for obvious problems like being damaged in transit or having checks or x-marks instead of filled-out ovals, inks other than black or blue, etc. These are pulled and duplicated in a process similar to the one at my site, except that they have stocks of preprinted ballots for each precinct that can be used. The ballots that look OK are batched in numbered boxes for tallying either there or at the site where I observed. The truck carrying the ballots between the two sites had a Democrat and a Republican observer along for the ride to make sure that the truck did not gain or lose any ballots along the way.

Our site was the last stop for doing ballot duplicates, and had no pre-printed ballots available. We had blank ballot card stock with a tear-off piece at the top. This was numbered with a dated counter that was crotchety unless it was held and pressed just right. It spoiled quite a few of the blank stock ballots, and each spoiled and successfully numbered ballot was recorded in a handwritten log book. After 100 pieces of blank stock were successfully numbered, our duplicators got their first batch of non-machine readable ballots. These were stamped with a red ORIGINAL stamp, red being ignored by the machine. The computer had a template for each precinct, and an associated numbered ballot style. The duplicators went through the originals one by one, and printed out a blank ballot in the correct style. (There was an annoying default—two ballots were printed, one in English and one in Chinese. Quite a few of the laboriously numbered blank stock pieces were ruined before someone figured out how to change the default setting to English only without messing anything else up. The HP printer printed both sides in the correct orientation simultaneously—thank heavens for one small favor.) In another log book, the precinct, batch and ballot style were noted along with each number, including spoiled ones that were printed incorrectly and the unused Chinese ones. The torn-off number stubs and the spoiled ballots were saved separately.

After lunch, our duplicators were ready to begin filling out the duplicate ballots. One had the stack of originals, and the other the stack of printed blank ballots. They first made sure they had the right ballot for the precinct, and the blank ballot was marked at the top with the precinct number and batch number. The original ballot was marked with the ballot tally number assigned by the dated incremental stamper, all in red ink. The information was also entered by hand in the log book. The original ballot choices were then read off by one person, as the second person marked the new ballot with a black marker. “Skip” meant that the voter had not made a choice for that position. I commented that it was a good thing that there were no candidates nicknamed “Skip,” and one of the poll workers said that last year there had been! Then they traded ballots, and the one with the duplicate ballot read off the names again as the other worker double-checked the original ballot. They were pretty good—they filled out only three of 70 wrong. These were marked as spoiled and so logged in the book, along with one original that had no blank mate and two blank ballots that didn’t go with an original.

More blank stock was numbered and printed. One more glitch—logging in again resulted in the default being landscape instead of portrait, which one of the supervisors was able to fix. After a great deal of work, 70 duplicates were sent to be batched for counting, and the originals, spoiled ballots and numbered tear-offs boxed up also. The rest of the shift was devoted to printing blanks for the next set of 100 ballots to be duplicated the next day.

These seventy ballots were from just the first three hours. Our duplicators sent word back to the first site that the workers there should be more stringent in their criteria for pulling ballots, as they had pre-printed stock there which required far less work for the duplicators. With most of these ballots, there was no immediately obvious reason why they should have been rejected by the machines. On some of them, the ovals were filled out sloppily so that the ballot mark overflowed the oval—probably by someone with Parkinson’s or an injury to their writing hand. The first site had caught all the write-ins and checked or x-marked ballots—I didn’t see any of these. On a couple, amateur cartoonists had taken advantage of the blank spaces on the back which wasn’t noticed on the first examination. We don’t disenfranchise people for being inconsiderate assholes in our state, but I can certainly see why our heavily burdened elections employees might be tempted to.

What I saw was people working hard and being completely open and transparent about what they were doing. This is, of course, the exact opposite of the status of the software that counts the votes, which is proprietary. Granted, the recount last year had a manual count of the gubernatorial race which was in 0.01% of the two machine counts, which indicates that none of the known back doors (three separate Access databases and the ability to cast negative votes, for starters) were actually used here. That is still no excuse whatsoever for their existence. Diebold proudly boasts on its website of its open source software for ATM machines—it seems that banks don’t like the idea of buying a pig in a poke to count their money. Nevads’s gaming commission likewise requires complete disclosure of all software and hardware design plans for its slot machines. Surely we should be demanding the same transparency for the software that counts our votes.

..to you for your contribution to making sure all votes count.

And thanks for telling your tale. At least we know some parts of the system still work as intended.