In addition to the lawsuit, Amico asked the state to investigate irregularities in the election.

The problem: Georgians cast nearly 4 million ballots on Election Day, but about 160,000 of them showed no vote cast in the lieutenant governor race, about 4.3 percent of ballots. To election experts, this so-called “undervote” rate — when a race is left blank — is evidence either that Georgia voters were unusually apathetic about their lieutenant governor, or that something went wrong.

It’s normal for 1 to 2 percent of voters to skip a race or races on a ballot, election experts say. But besides being more than double that normal rate, the number of uncast votes in the lieutenant governor race also exceeded Duncan’s margin of victory over Amico, which was just 123,172 votes.

The puzzling numbers call new attention to Georgia’s paperless, touchscreen voting machines, which drew lawsuits in 2017 from election-integrity groups that say the machines are not secure and want the state to switch to paper ballots that can be audited. Those lawsuits are ongoing, but after the midterm elections one of the groups, the Coalition for Good Governance, filed a second lawsuit with two Georgia voters and the losing Libertarian candidate for Georgia secretary of state, to invalidate the lieutenant governor results and conduct a forensic examination of the voting machines.

The lawsuit cited the abnormal undervote numbers as well as complaints from three voters who signed affidavits saying the lieutenant governor’s race either didn’t show up on their onscreen ballots or displayed oddly on the touchscreen voting machines. A state judge last month dismissed the suit on grounds that the complaints were limited to specific precincts, and any missing votes in those precincts would not be enough to alter the outcome of the race. The plaintiffs plan to appeal.

In her letter to interim Secretary of State Robyn Crittenden in November requesting an investigation, Amico said the high undervote rate indicates “serious inadequacies” in the administration of the election and possible problems with the voting machines. But Crittenden rebuffed Amico’s request, saying undervotes alone were not sufficient evidence of a problem. Amico expressed amazement at the state’s lack of interest in looking into what happened in her race.

“If this were an accounting department that found a missing $100,000 instead of 100,000 votes, there would be an investigation,” she told POLITICO. Amico, who is not a party to the lawsuit, did not call for the results to be thrown out.

“I’m literally seeking to make sure this never happens to any candidate from any party ever again,” she said.

The Georgia Secretary of State’s office did not respond to repeated requests for comment.

Longtime voting machine expert Douglas Jones said Amico and the plaintiffs ran into a common roadblock for people seeking to uncover the facts behind unusual election results.

“It’s the Catch-22 that’s been around forever where you can’t get any legal support [to investigate] unless you’ve got proof that something went wrong. But the evidence you need to get that proof is hidden from you because there is no access to the material you need to prove it,” said Jones, a professor of computer science at the University of Iowa.

Voters intentionally or unintentionally leave races blank all the time — for example, if they don’t like any of the choices in a race or they overlook the race on the ballot. And it’s not unusual for races that appear lower on a ballot to have higher undervote rates than those at top, because voters often deem down-ballot races to be less significant.

But election experts who filed affidavits in the lawsuit note that the undervote rate in Duncan and Amico’s race was the highest of any Georgia lieutenant governor race since 2002, which traditionally had rates around 1 percent. It was also four times the rate of the nine other statewide races on Georgia’s ballot last year: The only other race with a marginally high number of undervotes was the contest for agriculture commissioner, which had a rate of just 2.4 percent.

And the experts point to a telling difference in results based on the method voters used to cast their ballot: The undervote rate in the lieutenant governor’s race was just 1 percent on mail-in or absentee paper ballots, compared with 4.5 percent on ballots cast on touchscreens, according to Christopher Brill, a senior data analyst with TargetSmart, who analyzed the numbers for the plaintiffs.

The difference in undervotes between paper ballots and machine ballots is a sign that something was wrong either with the digital ballot layout — which may have caused voters to miss the race on screen — or with the machines, Jones said.

“The absentee ballots provide strong confirmation that this is unusual, because people who voted on paper had an opinion in this race” and wanted to vote in it, Jones told POLITICO. He said the data contradict claims by state election officials that the undervote rate was high because voters simply weren’t interested in the race.

Furthermore, the plaintiffs produced a report over the weekend showing that the high undervote rates on touchscreen machines were concentrated in precincts that are predominantly African-American. The rates in such precincts “are far greater than the undervote rates in non-African American neighborhoods regardless of whether those neighborhoods lean Democratic or Republican,” they wrote in their report. “The undervote problem did not happen at the same exaggerated levels in many primarily White neighborhoods that overwhelmingly voted for Stacey Abrams and other Democrats, rebutting the argument that the difference can be explained by party-driven voter behavior.”

Crittenden, in her denial of Amico’s request for an investigation, said the state did “parallel testing” of voting machines on Election Day, using sample machines and the same code used on machines in counties. These tests have “always shown 100% accuracy,” she wrote in her response.

But experts say parallel testing is not a reliable indication of Election Day problems. If voters overlooked a race because of confusing or poor ballot design, this probably won’t show up in testing, which is often done with a script instead of conditions that emulate real voting. And someone who wants to subvert a machine can design malicious code that senses if a machine is in test mode and only acts during a real election.

“It reminds me of the Wizard of Oz saying, ‘Don’t look behind that curtain, ignore that man,’” Jones said of the state’s refusal to investigate. “There is evidence that something is anomalous in this race, and the state’s response is ‘No, everything is fine. Please don’t look, don’t ask these questions.’”

Chip Lake, a campaign adviser for Duncan, told POLITICO that Amico and the lawsuit plaintiffs are guessing at a cause for the undervotes. He said that if the problem was a machine malfunction, it would have shown up in every race, not just one.

“Why was it only in that race, and why did [the high undervote rate] spread over 159 counties?” he said.

Instead, he suggested, the digital ballot layout confused voters. On paper ballots, the governor’s race appeared above the one for lieutenant governor. On voting machines, they were side by side. Lake suspects that voters either overlooked the latter race on the touchscreens or mistakenly thought the two races were linked — perhaps believing that if they cast a vote in the gubernatorial race, this would automatically cast a vote for the candidate from the same party in the lieutenant governor’s race.

But Sara Tindall Ghazal, director of the voter protection program for the Democratic Party of Georgia, said it’s unlikely voters overlooked the lieutenant governor race, because each was clearly marked and set off in a different box. If voters were expecting the races to be linked, she said, they would have noticed if no “X” appeared in the lieutenant governor race when they selected a candidate for governor.

Ghazal points to another possible culprit: In Election Day calls to the Democratic Party’s hotline and in affidavits filed with the lawsuit, she said numerous voters reported that the lieutenant governor’s race was missing from their ballot and showed up only on a review screen at the end of the selection process. Some said they could then scroll back and cast a vote for lieutenant governor — a process other voters might have missed. Asked how many voters complained about this, Ghazal said she didn’t know.

One voter, Chris Ramirez, who signed an affidavit in the lawsuit, said that only Duncan’s name appeared in the lieutenant governor race on page one of his touchscreen ballot. Amico’s name was on the next page, with no header or context indicating why it was there, he wrote in his account. Although he was able to select her name on that page, he wasn’t sure the machine recorded his vote.

Another voter, Ronica Johnson, told POLITICO she voted in the lieutenant governor race, but the review screen indicated she’d left it blank. She scrolled back through the ballot to make her selection again, but was so concerned the machine might not record her vote, that she reported the problem to a poll worker. She said the poll worker dismissed her concern.

A third voter, mentioned in the lawsuit but not identified, tried to go back to vote in the lieutenant governor race after seeing it was blank on her review screen, but when she touched the box next to a candidate’s name to make her selection in the race, the machine immediately jumped to a page indicating she had just cast her ballot, though she had not touched the “cast ballot” button. Her account of what occurred was supported by a polling place manager, who called the Democratic hotline. Another caller at the hotline said the lieutenant governor race never appeared onscreen until the voter reached the review screen at the end of the ballot.

Such problems might not have been limited to the lieutenant governor’s race, even without similar complaints from voters in other races, experts say.

“Any sign that the machines are failing certainly is evidence that you can’t trust any results from the machines,” says Walter Mebane a political scientist and statistician at the University of Michigan, who is not associated with the lawsuit. "[I]f one has doubts about the machines, then basically nothing is trustworthy from them.”

High undervote rates have posed problems before with paperless electronic voting machines. In 2006 in Sarasota County, Fla., for example, more than 18,000 ballots cast on paperless touchscreen machines recorded no vote in a hotly contested congressional race, an undervote rate of 13 percent. The victor, Republican Vern Buchanan, won by fewer than 400 votes.

Like Georgia, Florida election officials insisted voters had intended to leave the race blank. But voters in 19 precincts hadcomplained to poll workers on Election Day that the touchscreens were failing to register their touch — a common symptom of calibration issues on touchscreen machines.

That congressional election came four years after a baffling village council race in Palm Beach County was decided by a margin of four votes — after 78 voters failed to record any selection at all on the touchscreen machines. In that case, the race was the only one on the ballot, making it especially odd that so many voters would take the trouble to go to the poll and then intentionally cast a blank ballot.

Georgia is one of five states in the U.S. — alongside South Carolina, Louisiana, New Jersey and Delaware — that still rely entirely on paperless electronic voting machines, though the state plans to replace its machines with ones that use paper ballots. In the absence of paper ballots, the best way to investigate if something went wrong with the current voting machines in Georgia is to examine the code on them, experts say.

In the lawsuit seeking an investigation, Georgia Superior Court Judge Adele Grubbs at first agreed to let the plaintiffs examine some of the voting machines in three counties. But she specified that they could examine only the memory and not files and data used to program machines. The activists agreed even though they felt memory alone wouldn’t reveal if the machines had been misprogrammed or maliciously altered. Still, the counties didn’t comply with the order, according to plaintiffs. And when they complained about that to Grubbs, she ignored their argument that the counties weren’t in compliance and ruled to dismiss the case.

“It strikes me this is a case where on its face it looks like there was a problem, and any sensible jurist, it would seem to me, would want to know if any malfunction occurred in the [machines],” said election expert Henry Brady, dean of the Goldman School of Public Policy at the University of California at Berkeley, who is not involved in the lawsuit. "[T]he requests made by the plaintiffs were quite reasonable.”

Amico said she’s not dropping the issue. She spoke last month to a committee convened by the Georgia Democratic Caucus to look into a number of issues around the November midterms, including undervotes.

“If the Republicans have zero intellectual curiosity about what caused this kind of undervote rate to exclusively happen on touchscreen voting machines,” she told POLITICO, “we all need to ask ourselves what could their motivation be for such an egregious lack of curiosity.”