one candidate had more votes than her whole party
Schaerbeek, on the north side of Brussels, 18 May 2003, half past eleven at night. Belgium has voted for a new federal parliament and the canton's totals have come off the machines. On the sheet a number breaks a rule. A candidate has more preference votes than her entire list received, which cannot happen: a preference vote for a candidate is also a vote for the list, so the list can never be smaller than one of its names. Someone phones the college of experts, the seven people parliament appoints to watch the automated vote.
The candidate is Maria Vindevoghel. Born in Waregem in 1957, she cleaned aircraft for Sabena, the Belgian national airline, until it went bankrupt in 2001 and she became the spokeswoman for the people it left behind. In 2003 she ran her own list in Brussels-Halle-Vilvoorde. It was called MARIA, for Movement for Action, Resistance and an Independent Alternative, and across the whole constituency it was heading for 4,298 preference votes and 0.51 percent, far under the threshold. Except in one booth in Schaerbeek, where according to the PC she was bigger than her own party.
The surplus was 4,096 exactly. Written in binary that is a one followed by twelve zeroes: a single bit, the thirteenth, reading one where it should read zero.
The investigation starts the next morning, 19 May, with the canton president, officials from the interior ministry, people from Steria, the software firm, and from Bureau Van Dijk, which vetted the system, and the experts. The diskettes from each polling station are read again and the problem shrinks to a single booth. The experts ask for that booth's ballot box. Belgian voters in 2003 did not mark paper. They pointed a light pen at a screen, the machine wrote the choice onto a magnetic card, and the card went into an electronic urn that read it on the way in. The cards are still in the box. They are run through again, the fresh totals are merged with the original diskettes from every other station, and the new sheet is clean. The 4,096 votes are gone and nothing else has moved.
On 22 May the experts and Steria agree on the likely story, and the same day they test the PC that sat beside the president of the polling station all Sunday. The memory passes. They close the booth again in simulation and cast test votes for only the candidate in question. On 23 May they try once more. The incident never comes back.
Then they go through the source code, published before the election, around the counters. What they find is a program that is almost nothing. All the election data lives in one global structure in the PC's RAM. The counters are set to zero when the booth opens. Through the day each vote sits in a buffer as a copy of the card, and only at closing does the program walk the buffer and add the votes up. There is no double count and no cross-check, and no instruction anywhere that could have added 4,096 to anything. If a bit flipped, the report says, it did so between the opening of the booth and the close, which may mean her counter spent much of that Sunday as a zero with a one already lodged in it, waiting for the evening count to start from 4,096.
The report, signed by seven names and filed with the Chamber, settles on this: since no error was found in the software, the error was very probably caused by a spontaneous and random inversion of one binary position, a phenomenon, it adds, abundantly described in the specialist literature. That is the whole explanation. No cause is named. The word cosmic does not appear.
Other people supplied it. By 30 July 2003 a French-language tech site was running the story under a headline about Belgian elections distorted by a cosmic ray. The mechanism is real. A particle arrives from outside the solar system, hits the top of the atmosphere, and sends a shower of secondary particles down through the air; one of them passes through a memory cell in a PC in Schaerbeek and deposits enough charge to tip it. Chip designers call that a single-event upset. Bharat Bhuva, an engineer at Vanderbilt, used the Belgian case at the AAAS meeting in Boston in 2017 and said in the same breath that it was never confirmed, only arrived at by elimination. David Glaude of PourEVA, the Belgian group that campaigns against electronic voting, put it less kindly. A spontaneous bit inversion, he said, is "the last explanation" when no bug turns up.
Either way the candidate's part is the same. Maria Vindevoghel lost in 2003 by the margin everyone expected. She went to work for the ACV union, and in 2019 she was elected to the Chamber of Representatives at the head of the PVDA-PTB list for Brussels, sixteen years after a machine in Schaerbeek reported, for one night, that she was impossible. The official report never gives her name. It calls her a candidate of a specific list.
Tintoretto painted the older version of this in Venice around 1575. Jupiter wants his son by a mortal woman to live forever, so he carries the infant Hercules down to the sleeping Juno and sets him at her breast. She wakes and pulls away. The milk that sprays upward becomes the Milky Way, and the drops that fall become lilies. In the canvas at the National Gallery in London the lilies are missing, cut off with a third of the bottom before 1727; what is left is the baby, the hand thrown up, two jets of milk turning into stars, and Juno's peacocks watching with all their eyes. A galaxy, made in one movement by a mother who did not mean to.
Somewhere tonight a counter is reading 4,096 too high. A follower count, a step total, a balance. Nothing is wrong with it in any way that can be found, no rule says a follower count cannot exceed anything, and nobody phones anyone at half past eleven. The person it belongs to is looking at it right now and calling it a sign.
& take note of coincidence, think they reveal the penmanship
Schaerbeek got caught because the number tripped the one check the system had. The experts had a week, a ballot box and the published source, and the best they could do was a bit from nowhere. Everyone else gets their number and keeps it.
Nobody recounts the ones that look right.
The painting: Jacopo Tintoretto, The Origin of the Milky Way (c. 1575), National Gallery, London. Public domain.
Sources
- College of experts' report on the 18 May 2003 elections, DOC 51 0001/2: alert at 23:30, recount from the magnetic cards, tests on 22 and 23 May, the 13th binary position, the source code findings, "inversion spontanée et aléatoire"
- Maria Vindevoghel: Sabena, the MARIA list, 4,298 votes and 0.51%, the Schaerbeek error, elected in 2019
- Electronic voting in Belgium: light pen and magnetic card, the 4,096 votes, a cosmic ray offered as one explanation
- Netastuces, 30 July 2003: the cosmic ray headline
- Vice: Bharat Bhuva at AAAS 2017, never confirmed, and David Glaude's line
- The Origin of the Milky Way, c. 1575, National Gallery: the myth, the peacocks and eagle, the trimmed bottom third