← All posts

No paper, no recount: what Sarasota's 18,000 blank votes teach

A congressional race ended by 369 votes. In the same county, on the same machines, 18,000 ballots recorded no choice at all — and there was nothing left to check.

It is just past 2 a.m. on November 8, 2006, and in a windowless room in Tallahassee, a number on a screen has stopped moving. Florida's 13th Congressional District — a stretch of Gulf Coast communities running through Sarasota and Charlotte counties — has just produced one of the closest U.S. House results in decades. Republican Vern Buchanan leads Democrat Christine Jennings by 369 votes out of roughly 237,000 cast across the district.

Three hundred and sixty-nine.

That is a margin of 0.15 percent. Any serious recount would need to examine every contested ballot. And here is the problem: in Sarasota County — the county that contributes the overwhelming majority of votes to this district — roughly 18,000 ballots recorded no vote at all in the congressional race. Not a spoiled ballot. Not a protest. Just... nothing.

Eighteen thousand, four hundred and twelve undervotes. A margin of 369.

One of those numbers is fifty times larger than the other. And because the machines that produced the larger number left no paper behind, there was nothing to recount.

What "undervote" actually means — and why 13% is the wrong number

Every election produces some undervotes. Voters sometimes skip a race they don't care about. That is normal. In a typical congressional contest, the undervote rate runs somewhere between one and three percent.

In Sarasota County's CD-13 race in November 2006, it was 13.9 percent.

Statewide, in the same election, the congressional undervote rate averaged roughly 2.5 percent. In neighboring counties that used optical-scan paper ballots — where voters mark a physical sheet that is then scanned — the undervote in the same race was under 3 percent.

That gap is not a statistical quirk. It is a question mark the size of a congressional seat.

Sarasota County used paperless ES&S iVotronic touchscreen machines — direct-recording electronic systems, or DREs. A voter walked in, touched the screen, saw a summary of their selections, and pressed "Cast Ballot." The machine beeped. A counter incremented somewhere in flash memory. And that was it. No paper stub. No printed receipt. No independently verifiable record of what the voter had chosen.

The machine's log said 18,412 voters cast ballots without selecting a candidate in the congressional race. It could not tell you whether those voters deliberately skipped the race, accidentally skipped it, or tried to vote and had their choice fail to register. It could not tell you, because it had not been designed to tell you. The only artifact of each transaction was the number stored on a memory card.

The GAO looked. It could not find the answer either.

Congress eventually asked the U.S. Government Accountability Office to investigate. The GAO tested the systems. Its investigators put the iVotronic machines through their paces, looked for evidence of a software malfunction, and tried to determine whether a defect in the machines could have caused the anomalous undervote rate.

Their conclusion, in GAO-08-97T, was not reassuring — not because they found fraud or malfunction, but because they could not rule it out. The tests did not find a machine malfunction that caused the undervote. But they also noted, explicitly, that a voter-verified paper trail could have provided independent confirmation that the touchscreens recorded votes correctly.

Read that again: the government's own auditors said the system lacked the basic artifact needed to verify its own output.

A voting system that cannot be independently verified is not a voting system. It is a trust exercise.

The GAO did not find fraud. The lack of paper meant they also could not find that there wasn't any. The two findings are not the same thing, and conflating them is exactly how opacity defeats accountability.

The recount that couldn't happen

Florida law allows a manual recount when the margin falls within 0.5 percent of votes cast. This race cleared that threshold easily — 369 votes on 237,000 cast is well inside it. Christine Jennings requested a recount.

There was nothing to recount.

A manual recount requires physical ballots — paper you can hold, examine, and count again. The iVotronic machines had stored each vote in electronic memory. You could ask the machines to re-tally, but that just re-reads the same data from the same storage that produced the original number. It is not an independent check; it is running the same process twice and expecting a different answer.

The courts looked at this problem and found themselves similarly stuck. Florida's courts and the U.S. House Administration Committee both investigated. The seat was ultimately certified for Buchanan, and he was seated. Jennings eventually dropped her challenge in 2008. No one ever established what had happened to those 18,000 votes.

The Sarasota race did not end in a courtroom finding of fraud. It ended in an institutional shrug, because the evidence needed to resolve the question had never existed.

That is a different kind of failure — quieter than fraud, harder to fix, and in some ways more dangerous, because it leaves the question permanently open.

Why the hand count is not a sufficient answer

You might reasonably ask: even if DREs are problematic, doesn't a good old-fashioned hand count solve this? In Florida 2000, we all watched people squinting at hanging chads and arguing about voter intent. In Georgia 2020, election workers hand-tallied all five million presidential ballots across 159 counties and confirmed the machine result to within about a tenth of a percent. In Colorado, the state pioneered risk-limiting audits that apply statistical rigor to how many ballots you need to examine to have confidence in a result.

These are genuine advances. But they all depend on one thing that Sarasota 2006 didn't have: a physical ballot the voter actually marked.

Optical-scan systems — where you fill in bubbles on a paper form, which is then scanned — give you that. The paper is the vote. If the scanner miscounts, as happened in Windham, New Hampshire, where a fold crease through a ballot target was misread as a marked oval, you can go back to the physical ballots and count them by hand. The Windham forensic audit, ordered by the New Hampshire legislature, traced exactly what had gone wrong: a folding machine had creased ballots along vote targets, the scanner read the crease as a mark, and no fraud was needed to produce a significant miscount. But the error was findable, because the paper existed.

With a paperless DRE, there is no going back. The ballot the voter saw on screen was never written to anything a human can read. The recount is the original count. The audit audits nothing but the machine's own memory.

This is not a subtle distinction. It is the entire ballgame.

The standard the German court set — and what Sarasota couldn't meet

In 2009, Germany's Federal Constitutional Court ruled on the use of voting computers in the 2005 federal election. Its reasoning was precise. The Court held that the use of electronic voting machines requires that the essential steps of the voting and of the determination of the result can be examined by the citizen reliably and without any specialist knowledge. This flows, the Court said, from the constitutional principle of the public nature of elections.

The machines failed that standard. Because votes were stored only in electronic memory with no independently verifiable record a voter could check, ordinary citizens — not cryptographers, not government auditors, not election lawyers — could not examine the count for themselves. The machines were declared unconstitutional.

The Court did not invalidate the 2005 election, because no malfunction was proven. But it drew a line: a democracy cannot rely on a counting system whose correctness the public must simply trust.

Sarasota's iVotronic machines would not have passed that standard. The GAO said as much in plainer language: there was no independent confirmation available. When 18,000 votes vanish into a machine's memory, a constitutional court standard of "ordinary citizens can check this" is not being met.

The Dutch Election Process Advisory Commission reached the same conclusion in its 2007 report, Stemmen met vertrouwen — "Voting with confidence." Its core finding: there are no secrets in the election process, and questions must be answerable, the answers checkable and verifiable. Any electronic method is acceptable only if it produces a paper vote the voter can check. The Netherlands subsequently returned to paper ballots and manual counting.

Both countries — one through a constitutional ruling, one through a government commission — arrived at the same answer. The standard is not "did anyone prove something went wrong?" The standard is "can the public verify, independently, that nothing went wrong?"

Sarasota answered that question in November 2006. The answer was no.

What verifiability actually requires

Here is what a verifiable election in Sarasota County in 2006 would have looked like, in practical terms.

A voter touches the screen. The machine prints a paper record — human-readable, showing the choices made. The voter inspects that record through a window on the machine. If it reflects their choices, they confirm and the ballot drops into a sealed box. If it doesn't, they can void it and try again.

This is called a voter-verified paper audit trail, or VVPAT. It is not a theoretical concept. It is hardware that existed and was available. The question was whether jurisdictions required it.

Sarasota County, in 2006, did not.

With a VVPAT, the congressional recount would have been a real recount. Workers could have examined 237,000 paper records and tallied the CD-13 contest by hand. They might have found that some voters genuinely skipped the race. They might have found that a subset of records showed no congressional choice despite the voter having made one on screen — evidence of a machine defect. They would have been able to tell you something. Instead, they had nothing.

The fix is not complicated in concept. It is — was — a printer attached to a touchscreen machine, producing a paper artifact that serves as the official ballot. What made Sarasota possible was not some exotic technical failure. It was the choice to deploy a counting system with no independent output at all.

The gap this case names

The Sarasota undervote has never been explained. It may have been voters who truly skipped a down-ballot race. It may have been a user-interface problem that caused some voters to miss the congressional contest when navigating a long ballot. It may have been something else. We do not know, and we cannot know, because the evidence was never created.

That is the lesson worth carrying. Not "fraud happened in Sarasota." There is no basis for that claim. The lesson is starker: a result that cannot be independently verified is not a verified result. The 369-vote margin and the 18,000-vote gap coexist in the same county, on the same election day, with no way to reconcile them.

Officials certified the result. Courts declined to overturn it. A congressional seat was awarded. None of that tells you whether 18,000 people tried to vote in a race and had their choices silently lost.

The Secretary of State's certification and the court's ruling are institutional resolutions. They are not answers.

The difference between "this race was resolved" and "this race was verified" is exactly the gap that better election systems are built to close.

What would close it: voter-marked paper ballots, mandatory voter-verified paper audit trails on any touchscreen system, routine risk-limiting audits that compare machine totals against physical ballots, and precinct-level results published instantly and in machine-readable form so any observer can check the numbers from the outside. Not because fraud is assumed. Because verification is what makes the absence of fraud provable — to you, not just to officials.

Sarasota 2006 is a paperless machine's clearest argument against itself. It did not produce a scandal. It produced a permanent question mark — which is worse.

What other verification gaps look like this? See how common this gap is across the world. Want the two-minute version? Read it here. For a deeper look at what "no paper to recount" means structurally, see our gap profile.

Sources

See all postsExplore the gaps