State officials, the Republican Party and the Trump campaign contested the recounts in all three states. And when Stein sought access to the software code used in Wisconsin’s voting machines — something state law permits for recount petitioners — the vendors who made the voting machines waged a protracted legal fight that has left Stein’s computer experts still waiting to see the code, four years later.

In the end, courts dismissed the recount effort in Pennsylvania, and halted the one in Michigan three days after it began. And once Wisconsin’s was done, legislators there amended the state’s law to ensure that no one like Stein, who had finished a distant fourth, could petition for a recount again. Now only candidates who lose by 1 percentage point or less can do so.

Stein’s experience is far from unique: Time and again, people seeking answers about perplexing results or election anomalies encounter obstacles. Often the only examination occurs if parties or candidates request a recount — which means voters are largely at the mercy of partisan actors to even raise the questions.

The truth is that two decades after the Florida 2000 election debacle created a rift in the country, and four years after Russian interference in the 2016 election profoundly deepened that divide, the U.S. lacks satisfactory, uniform mechanisms for resolving questions about elections and verifying results.

The U.S. lags behind most other countries in this regard, said David Carroll, director of the Democracy Program for the Carter Center, which has monitored foreign elections since 1989 — even if many reasons exist to have overall confidence in the American electoral process.

“The United States has many areas where we don’t meet core international standards,” he said.

Almost every other democracy, he said, has uniform, nationwide election procedures; a central, independent election commission to oversee them and conduct investigations; a single, national election system; and an unvarying system for resolving disputes. “Whereas in the United States, there [are] 50 different election processes, 50 different sets of state laws,” Carroll said.

Also uncommon elsewhere: The U.S. places large portions of its electoral process under the control of elected state and local politicians, raising the potential for partisan interference.

In the absence of mechanisms that support election integrity, the U.S. has largely had to rely on “public trust … and acceptance of the results,” Carroll said. This has mostly worked out, until now.

But this presidential election may be the one where that deficit finally catches up with us. With Trump proclaiming again that the election is “rigged” and raising concerns that he may not accept the outcome if he doesn’t win, any unexplained or uninvestigated anomalies could leave millions of Americans distrusting the outcome, thereby deepening the divides that already exist and potentially inspiring violence. Before his surprise win four years ago, Trump seized on machine malfunctions in states like Pennsylvania as evidence that the election should not be trusted.

Dan Coats, Trump’s former director of national intelligence, seemed to acknowledge the lack of integrity mechanisms in an op-ed for The New York Times in September, when he proposed that Congress establish an independent, bipartisan commission to combat efforts to undermine trust in the current election. The commission “would not circumvent existing electoral reporting systems,” he wrote. “[It] would monitor those mechanisms and confirm for the public that the laws and regulations governing them have been scrupulously and expeditiously followed … without political prejudice and without regard to political interests of either party.”

Beyond Coats’ immediate solution for this election, election experts say long-term trust could be achieved with a federal law mandating hand-marked paper ballots in every state and rigorous post-election audits. Those steps could help determine with high probability that the correct candidate has won and also uncover systemic problems with machines and election processes. But such efforts have repeatedly failed on Capitol Hill, most recently with legislation introduced by Sens. Ron Wyden (D-Ore.) and Amy Klobuchar (D-Minn.). Two of their bills — the PAVE Act and the SAFE Act — failed to advance in the Republican-dominated Senate.

Wyden said he sees no reason to believe the results won’t be trustworthy this year. “But there are plenty of common-sense steps that nonpartisan experts agree Congress and states can take to improve confidence in our elections,” he told POLITICO in a statement. In addition to paper ballots and audits, these would include standards for ballot designs that don’t confuse voters and cause them to make mistakes, as well as “creating uniform, science-based clear standards for ballot signature verification, and ensuring states know what to do next if audits uncover irregularities after an election.”

Edward Foley, a constitutional and election law expert at Ohio State University, said the fact that the U.S. system is not perfect or ideal doesn’t mean it’s inadequate. But its success rests on two questions: Does it produce the correct winner? And is each voter fully able to participate and have their votes accurately counted?

If the answer to the first is yes, “then the system … is working to achieve its intended purpose, even if there are significant problems along the way,” he told POLITICO. “But at the level of an individual voter, are they able to participate in the system successfully without personal disenfranchisement? That's not always true, even if the system is achieving the right results."

Efforts to bolster election integrity don’t just ensure the correct outcome. They also help determine the rates of errors and voter disenfranchisement.

“You'd like to get it as close to zero as possible, even if it's never making a difference” in the results, Foley said.

But investigations into irregularities that disenfranchise voters seldom occur.

Investigations deficient or thwarted

One example of an investigation that remained incomplete: A probe by House Democratic staff found what it called "massive and unprecedented” irregularities in the 2004 election in Ohio that Democrat John Kerry lost narrowly to former President George W. Bush. Their report called for Congress to hold bipartisan hearings and explore creating national election standards that “would promote the guaranteed right to vote and would ensure that every vote counts.” But neither of those things occurred.

Another incomplete investigation involved a 2006 election in which Florida Republican Vern Buchanan won a U.S. House seat by fewer than 400 votes. More than 18,000 ballots in Sarasota County showed no votes cast in the House race — a gap known as an undervote. State and federal probes examined voters’ complaints that the paperless touchscreen voting machines may have failed to record their choice in the race, while county officials insisted that voters had intended to leave the race blank or had simply missed it on the ballot.

Florida’s Department of State conducted tests and concluded that the machines weren’t at fault, but experts called the tests flawed. More extensive testing by the Government Accountability Office “did not identify any problems” with the devices but also couldn’t say definitively that they hadn’t dropped votes, only that they were unlikely to be the cause. The questions could have been answered easily if the election had used paper ballots, which would have provided a permanent record of the voters’ selections.

Both of these were examples where investigations at least occurred. But other election problems get no investigation at all.

One example involved the sudden disappearance of 16,000 votes from Al Gore’s tally on the night of the 2000 election, after news networks had already projected him the winner in Florida over Bush.

The problem was Precinct 216 in Volusia County, where Gore’s total suddenly dropped to an impossible minus 16,022 votes. Logs examined later for the county’s tabulation system, made by Diebold Election Systems, indicated that someone had tabulated two voting machine memory cards from Precinct 216 instead of just one. The tabulator subtracted votes from Gore’s total after the second card was inserted. The county was able to determine the real tally from a hand recount of the paper optical-scan ballots, but why the votes disappeared is still unknown.

In internal emails that were later leaked online, Diebold employees speculated that the second card might have been corrupted, but also joked that it might have been an “unauthorized” rogue card introduced by someone intentionally. Either way, no one could locate the card to examine it, and the matter was dropped as the country’s attention focused on more prominent problems in Florida — hanging chads, Palm Beach County’s butterfly ballot and the Supreme Court’s intervention into the recount. Bush won Florida by 537 votes.

Four years later, intervention of another kind — this time by the Justice Department — thwarted a local investigation into mystery ballots in Maricopa County, Ariz., in an unusually striking example of how election probes get interrupted.

Anton Orlich, a far-right conservative, had defeated establishment Republican John McComish by four votes in a GOP state House primary. An automatic recount overturned the results and gave McComish the win by 13 votes, but 489 previously uncounted votes also materialized. Orlich appealed to friends in the state Senate to investigate, and they hired University of Iowa computer science professor Doug Jones.

Jones, who used to chair the board responsible for testing and certifying voting machines in Iowa, tested six Maricopa ballot-scanning machines and found big disparities in how they treated ballots marked with pencil or black pen. The scanners were “only marginally sensitive” to ink and overly sensitive to pencil, interpreting tiny lead specks and smudges as votes.

Jones believed that some of the scanners had either missed votes during the initial tally or erroneously counted stray pencil marks during the recount. Another possibility: Someone added votes to the ballots before the recount. Jones wanted to examine the ballots, but before he had a chance, FBI agents intervened and seized them.

The local U.S. attorney, who had decided to launch his own probe, announced 10 months later that the bureau had found no evidence of fraud. The additional votes, he said, were probably just due to a more accurate machine being used for the recount.

Jones says this may be the case, but the summary of the FBI report that he saw contained no information about how the bureau conducted its investigation or reached this conclusion.

“[I]t read like a whitewash job to me,” he said.

Probes blocked, evidence destroyed

Without mechanisms to support election integrity, many things can undermine it — even something as basic as public perceptions. Candidates who ask for recounts often get labeled sore losers, and election integrity activists who seek investigations of irregularities are often mocked as conspiracy theorists. (Stein’s critics accused her of pursuing her recount effort as a fundraising ploy, and Trump dismissed it as a “scam.”) Meanwhile, the public and the news media tend to lose interest in elections once the horse race is done, especially if the victory margin is wide — even if that allows systemic problems to go unaddressed.

Other obstacles include:

— Time. Deadlines to file recount requests or election challenges fall just days after the votes are tallied in some states. In presidential elections, vote certification and Electoral College deadlines restrict the time available for recounts. This “makes the likelihood of an efficiently-administered and properly-litigated recount effectively impossible in most states,” election law expert Maya Efrati wrote in a 2018 report by the election reform group Fair Vote.

— Resistance by voting machines vendors, who fight efforts by candidates and others to examine their systems. The vendors usually investigate their own machines, often blaming voters, poll workers or election staff for whatever went wrong. That happened in 2008, when a tabulation system made by Diebold (then calling itself Premier Election Solutions) dropped hundreds of votes in Ohio counties during the presidential election. The company faulted election workers and an antivirus program the counties had installed, before finally admitting the problem was a decadeold programming error in its software. (The discrepancy didn’t affect former President Barack Obama’s 260,000-vote victory in the state.)

— Missteps by election officials, even if the vast majority are honest, hard-working professionals. In 2004, two election officials in Cuyahoga County, Ohio, preparing for the recount between Kerry and Bush pre-selected ballots that they knew would support the original results. They were later convicted of negligent misconduct.

— Destroyed data. Election officials are legally required to retain federal election records 22 months after polls close. But Georgia election officials wiped a server clean in mid-2017, even though it was less than a year after the 2016 election and the data was central to a lawsuit filed by an election integrity group.

“If you don’t get to the court to get a protective order to stop the destruction of something before they’ve done it, it’s typically a pretty lost cause,” said Lowell Finley, chief counsel for California’s secretary of state until 2014 and a consultant on the Stein litigation. He said election officials violate the record preservation law because there’s little to stop them.

Similarly, 12 months after a 2016 congressional primary in Broward County in South Florida, election supervisor Brenda Snipes destroyed optical-scan ballots that were part of a lawsuit filed by losing Democratic candidate and law professor Tim Canova. County attorneys insisted that Snipes did nothing wrong because she retained digital copies of the ballots, but a judge later declared the destruction unlawful. Snipes kept her job despite the infraction, but later resigned amid criticism of her handling of the problem-plagued 2018 election.

— Records that never existed to begin with. In Humboldt County, Calif., during the 2008 presidential election, a Diebold tabulator inexplicably dropped 197 ballots from the tallies of the county’s scanned-in paper ballots. County officials discovered that the system’s logs had failed to record critical information that could have explained when and why the ballots disappeared — a lapse that violated the federal voting system guidelines that systems used in California are supposed to meet.

In 2006 in Ohio, a Franklin County programmer disabled a voting machine logging function that was supposed to track any changes made to software on the machines. He said the voting machine vendor, Election Systems and Software, advised him to do this because it would speed up the process of programming the machines before elections.

— Officials who refuse to investigate. This happened after the 2018 lieutenant governor’s race in Georgia between Republican Geoff Duncan and Democrat Sarah Riggs Amico. More than 160,000 ballots showed no votes cast in that race. Voters sometimes leave races blank for legitimate reasons, but a normal undervote rate is 1 to 2 percent of ballots cast. The rate in the Georgia race was 4.3 percent and exceeded Duncan’s margin of victory (123,000 votes). The undervotes were concentrated in predominantly Black precincts, and Georgia’s voting machines were paperless, leaving no backup ballots to determine the voters’ intentions. Election officials said the problem was a poor ballot design that caused voters to miss the race.

Amico asked the Republican acting secretary of state, Robyn Crittenden, to investigate. But Crittenden declined, saying undervotes alone weren’t evidence of a problem. Amico told POLITICO at the time: "If this were an accounting department that found a missing $100,000 instead of 100,000 votes, there would be an investigation.”

Finley said election officials generally dislike investigations because they eat up time and resources when officials are trying to certify election results. But they also hate them for another reason:

“For elections officials … the only thing that can come out of a thorough examination of what might have gone wrong in an election is that [it’s going to show] there was a mistake or some kind of malfeasance and the result of the election was incorrect. So they have a very strong interest in not letting that happen,” he said.

— Lack of thoroughness. In 2015 in Shelby County, Tenn., election staff missed the fact that their tabulation system had dropped more than 1,000 votes in a municipal election until a voter alerted them to it.

The voter, Bennie Smith, photographed the poll tape that had been posted outside his precinct, showing that people had cast 546 ballots there. (Poll tapes are receipts that poll workers print from voting machines at the end of an election, showing a tally for all votes cast on the machines.) But the county’s official tally showed only 330 ballots cast in Smith’s precinct. When an official looked into the matter, he discovered discrepancies in three additional precincts. The tabulation system had dropped or failed to record 1,001 ballots — all from predominantly Black precincts.

Election workers would have caught the error on their own if they had compared the poll tapes with tabulated results. But they compared tapes from only 10 percent of precincts; the precincts with the dropped votes weren’t among them.

Recounts and challenges

When other integrity mechanisms aren’t an option, recounts can at the very least help determine if the outcome was correct, but they tend to be politically charged and overly focused on results rather than uncovering systemic issues. And candidates and campaigns rarely pursue them unless they stand a chance of overturning the results.

Twenty-one states and the District of Columbia have automatic recounts if the margin is narrow, usually 1 percent or less. Forty-three states and D.C. let candidates or voters petition for a recount, but the requirements vary. In some states, the petitioner has to show probability that fraud occurred or that the number of ballots affected exceeds the margin of victory.

Even when candidates or voters meet those requirements, it doesn’t mean the recount will be done satisfactorily. The 2016 recounts illustrate this.

Experts who urged Stein to seek recounts in Wisconsin, Michigan and Pennsylvania did not claim evidence of fraud, only that it would be worthwhile to verify the results. One political science and statistics professor from the University of Michigan also pointed to a pattern in Wisconsin: the frequency at which Clinton’s vote percentages, and her combined percentages with Trump, ended in 0 or 5. When Russians rig an election, professor Walter Mebane Jr. wrote in The Washington Post, they often signal that they’ve accomplished the deed to whoever ordered the fraud by manipulating the results to end in a pattern of 0 or 5.

The Clinton campaign opted not to pursue recounts in the absence of evidence that voting machines were hacked or a chance of overturning the results, Marc Elias, the campaign’s general counsel, wrote at the time. Recounts rarely flip outcomes. Of 4,687 statewide elections conducted across the U.S. between 2000 and 2016, only 26 resulted in recounts, according to Fair Vote; only three of those changed the outcome. The margin in Michigan (0.23 percent) — the smallest among the three states Stein petitioned — was nearly double the largest margin ever overturned in a statewide recount (0.12 percent).

Stein argued that the recounts weren’t about changing results but about restoring voters’ confidence in the democratic process after a “deeply painful election.”

But both the Michigan and Wisconsin recounts proved to be problematic because of their state laws.

Wisconsin’s recount law lets counties recount ballots either by hand or by simply scanning them through the machines that originally counted them — a method that runs the risk of repeating previous errors. About 85 percent of ballots were cast in Wisconsin using optically scanned paper ballots, which would have been easy to recount by hand, but a dozen counties rescanned all of their ballots and about another dozen rescanned some of them.

The Wisconsin recount didn’t change Trump’s victory, but it did reveal that about 1 in 116 ballots had votes that had been miscounted or uncounted in the original tally. A lot of those were due to voter error, such as using gel pens or red ink that the scanners couldn’t read or failing to follow directions for adding a write-in candidate’s name.

The partial Michigan recount also uncovered irregularities and evidence of ballot mishandling. When officials opened bins of ballots to recount in Detroit, they discovered hundreds of ballots missing from about three dozen precincts. Workers had left them in the tabulator on election night instead of moving them to sealed bins, breaking their chain of custody. Officials also discovered that 392 of Detroit’s 662 precincts had fewer or more ballots than the number of people recorded as having voted. In most cases, the difference was just a few ballots, but combined they added up to 782 more ballots than voters.

Such discrepancies often occur for a simple reason: Poll workers fail to record all the voters in the precinct poll book or fail to record the number of provisional ballots they give voters who aren’t listed in that book. But Michigan has an unusual law that took the Stein team by surprise. Any precinct that has mismatched ballot and voter numbers or has ballots that are not placed under seal on election night is excluded from recount, and the original count for the precinct stands.

Rick Hasen, a professor of law and political science at the University of California, Irvine, calls the Michigan law “ridiculous” because it means a true recount can’t occur. It “would lead you to a less accurate count,” he said.

The 2016 recounts in Wisconsin and Michigan left Trump’s victories intact. But Maazel, who helped lead Stein’s legal fight, wrote afterward that the recounts “did not confirm the integrity or security of our voting system; they revealed its vulnerability.”

And Stein did help achieve at least one change in Pennsylvania. About 80 percent of the state’s voting machines were paperless in 2016, which meant there was no effective way to do a recount there or any other investigation to verify results. Stein's legal challenge there ended in a 2018 settlement in which the state agreed to replace all its paperless voting machines with ones that can be audited, and to mandate post-election audits. (Democratic Gov. Tom Wolf had announced a similar push months earlier.) However, some counties replaced them with machines that have their own integrity issues.

All the experts POLITICO spoke with said the U.S. electoral system needs changes, and they agreed on how that should occur — with paper ballots, mandatory robust audits and an independent, nonpartisan commission to oversee elections and provide a method of recourse when things go wrong. The question is: Are Americans ready to make those changes?

“We got the secret ballot in the 1890s because it suddenly became imperative that we needed a secret ballot,” Foley told POLITICO. “If suddenly it became salient to say we can’t really run an election without [stronger] audits, then that would be the new reform that would take hold like the secret ballot did. But we’re not there culturally. The average person doesn’t see the necessity of that.”