Voter Fraud Is Real. So Is Arithmetic.

Suppose investigators prove ten illegal votes occurred. Ten sounds alarming if the denominator is missing. Ten out of twenty ballots would be catastrophic; ten out of 158 million would describe a very different scale. The U.S. Election Assistance Commission reports that more than 158 million ballots were counted in the 2024 general election.^[344][345] That number does not tell us how many fraudulent ballots existed, but it gives us something equally necessary: a national election denominator against which properly defined evidence can be placed.

A fraud rate needs both pieces. A collection of prosecutions without a defensible denominator is not a national fraud rate, and a denominator without a defensible fraud numerator is not one either. The same discipline applies to the earlier stages of the pipeline: 1,000 suspicious registrations are not 1,000 fraudulent votes, 100 investigations are not 100 referrals, 50 referrals are not 50 charges, and 10 charges are not 10 convictions. One proven illegal ballot proves that an illegal ballot existed. It does not, by itself, establish that fraud was widespread or that an election result was wrong. That is the rule we will keep returning to throughout this article: fraud happening is not the same as enough fraud happening to change the outcome.

At the federal level, the Justice Department's Public Integrity Section supervises the nationwide investigation and prosecution of election crimes.^[348] Its Election Crimes Branch reviews major election crime investigations and proposed criminal charges, while U.S. Attorneys' Offices around the country prosecute cases in their federal districts.^[342] DOJ identifies federal election offenses including vote buying, multiple voting, fraudulent ballots or registrations, destruction or alteration of ballots, and malfeasance by election or postal officials.^[349]

That gives researchers several places to look: the Election Crimes Branch, DOJ press releases, individual U.S. Attorney offices and ultimately federal court records. But DOJ is tracking federal criminal enforcement. It is not publishing a census of every suspicious registration, rejected ballot, state-law referral or county prosecution in America.

The decentralization that makes a single national count difficult also means the primary records are usually public if you know where to look. For federal cases, start with the DOJ Election Crimes Branch and the relevant U.S. Attorney's Office, then check the federal court docket.^[342][348]

For state cases, start with the state's Secretary of State, election board or election agency and Attorney General, then check the relevant county prosecutor and state court system. Georgia's Secretary of State publishes election investigation and referral information,^[350][351] Michigan's Attorney General maintains election protection and enforcement information,^[352] and North Carolina's State Board of Elections maintains an Investigations Division with referred case summaries and annual case data.^[346]

Administrative investigations may also appear in election board meeting materials, referral reports, audits and formal investigation summaries. North Carolina is unusually helpful here because its Investigations Division publishes referred case summaries dating back to 2015.^[346] That is excellent transparency, but a referral in those records is still a referral, not automatically a conviction.

Now imagine trying to calculate a national fraud rate from these ingredients: one state publishes complaints, another publishes investigations, another publishes referrals, a county publishes prosecutions, a federal office publishes convictions, and some cases are discovered years after the election. Adding those numbers together would combine different units, and dividing the result by national turnout would produce a percentage with impressive decimal places and very little meaning.

A defensible national rate would require a consistently defined numerator, comparable detection methods, known coverage and a meaningful denominator, and the United States does not currently produce that dataset as one standardized government statistic. So throughout this article, when we can measure something precisely, we will; when we only have a minimum documented count, a referral count or an illustrative rate, we will call it exactly that. The spreadsheet does not get to outrank the evidence.

Iowa gives us one of the clearest demonstrations of why the stages matter. Before the 2024 election, the state was working from a list of 2,176 registered voters who had previously self-reported noncitizen status to the state Department of Transportation. That did not establish that all 2,176 were still noncitizens. Some people naturalize after obtaining a driver's license or otherwise have stale records.

After Iowa gained access to the federal SAVE system and conducted additional verification, the Secretary of State reported 277 confirmed noncitizens in the reviewed population.^[354] Further review found that 35 of those people cast ballots that were ultimately counted in the 2024 general election, while five attempted ballots were rejected. Twenty-two had registered in 2024 but did not vote. Iowa referred the cases for investigation.^[354]

The progression matters more than the headline: 2,176 potential matches became 277 confirmed noncitizens, and 35 counted ballots. The original list was useful for finding cases. It was not itself a list of 2,176 illegal voters.

If your statistic gets eighteen times smaller when somebody checks what the records actually mean, checking the records was probably an important step.

North Carolina found the same problem years earlier. Its audit of the 2016 election ultimately identified 41 ineligible noncitizens with legal immigration status who voted among nearly 4.8 million voters. But the same audit also reported that 34 voters initially flagged by the citizenship review subsequently provided proof of citizenship. The audit explicitly said those results highlighted why data matches alone were not sufficient to verify citizenship without follow-up investigation.^[347]

That can happen for mundane reasons. A person may have been a noncitizen when an older government record was created and later become a naturalized citizen. Records can also be incomplete or mismatched. NCSL's current review of citizenship list-maintenance systems explains that USCIS's SAVE system provides point-in-time status and was not originally developed as a voter-registration system. It is not a comprehensive citizenship registry, and additional information can sometimes be necessary to determine eligibility.^[357]

None of this means states should ignore citizenship mismatches. It means a mismatch is something to investigate rather than something to paste directly into a chyron labeled ILLEGAL VOTES.

Michigan gives us a recent example with a particularly clear denominator. After comparing motor-vehicle records with the state's Qualified Voter File, Michigan reported 15 people who appeared to be noncitizens and had cast ballots in the 2024 general election. The state put that figure at 0.00028% of more than 5.7 million ballots. Thirteen cases were referred to the Attorney General for potential prosecution, one person had died, and one case remained under investigation when the results were announced.^[355]

North Carolina's 2016 audit found 41 ineligible noncitizens with legal status among nearly 4.8 million voters, roughly 8.5 per million.^[347] Those are real findings from two different state processes. They are also not directly interchangeable: Michigan described credible or apparent cases and referrals, while North Carolina's older audit used its own verification process and population. Different methodology means we should resist the temptation to average them together and call the result America.

The useful conclusion is narrower. In audits where we have both a defined numerator and a large election denominator, detected noncitizen voting has represented a tiny fraction of turnout. That does not prove undetected cases equal zero. It tells us what those particular audits actually found.

This distinction is non-negotiable. A noncitizen can be a lawful permanent resident, a visa holder, a Canadian citizen living in the United States, another lawfully present foreign national, or an undocumented immigrant. Those categories have different immigration statuses even though none of them is a U.S. citizen.

North Carolina's 41 voters were specifically described as noncitizens with legal status, including green-card holders.^[347] Calling those 41 people 'undocumented immigrants' would not sharpen the statistic. It would make it false.

Our broader research has also located criminal cases involving both lawful noncitizens and people specifically identified in government records as unlawfully present. So the correct conclusion is not that undocumented voting never happens. It is that a dataset measuring noncitizens cannot automatically tell us how many of those noncitizens were undocumented.

We went looking for a defensible national number of undocumented immigrants who vote. The available government data do not support one. Inventing a conversion factor from 'noncitizen' to 'undocumented' would be Fraud Math wearing an immigration hat.

Administrative audits are not the only attempt to measure noncitizen voting. A 2014 paper by Jesse Richman, Gulshan Chattha and David Earnest used Cooperative Congressional Election Study survey data and concluded that noncitizen participation was low but nonzero, with estimates large enough that the authors argued it could have affected some election outcomes.^[359]

A 2015 response by Stephen Ansolabehere, Samantha Luks and Brian Schaffner challenged that conclusion on methodological grounds.^[360] Their argument is important beyond this particular political dispute: when the event being measured is extremely rare, even a tiny amount of classification error can swamp the signal. If a very small number of citizens are accidentally recorded as noncitizens in a large survey, those errors can become a substantial portion of the apparent noncitizen subgroup.

We do not need to settle an academic argument by declaring one paper forbidden. We need to understand what each method can support. Survey estimates can illuminate behavior but become fragile when classification errors are similar in size to the rare event being measured. Direct administrative audits have their own limitations, but they let us observe the verification funnel much more concretely.

The Brennan Center approached the problem differently after the 2016 election. Researchers interviewed election administrators representing 42 jurisdictions in twelve states, including jurisdictions with large noncitizen populations. Those officials oversaw 23.5 million votes and reported an estimated 30 incidents of suspected noncitizen voting referred for further investigation or prosecution.^[358]

That works out to roughly 1.3 suspected incidents per million votes in the jurisdictions studied. But notice every qualifier in that sentence. The incidents were suspected, not necessarily convicted. The 42 jurisdictions were not every jurisdiction in the country. And election officials acknowledged that their detection tools were imperfect.

It is therefore useful evidence about observed scale in those jurisdictions. It is not permission to multiply 1.3 by every election in American history and announce that we have solved voter fraud.

We can perform an illustrative calculation, but we should not pretend it is a national estimate. If the Brennan jurisdictions' observed suspected/referral rate were mechanically applied to the more than 158 million ballots counted in 2024, the arithmetic lands at roughly two hundred incidents.^[358][344] Applying Michigan's 2024 apparent-case rate produces roughly four hundred.^[355] Applying North Carolina's 2016 audited rate produces roughly thirteen hundred.^[347]

Those calculations suggest a scale measured in hundreds to low thousands rather than millions. But they do not establish that the true 2024 national number falls between those endpoints. The underlying datasets come from different years, states, populations, detection systems and definitions. One numerator counts suspected referrals; another counts administratively identified cases; another comes from an audit of legally present noncitizens.

So we will not publish '200–1,300 illegal noncitizen votes in 2024' as though we discovered a confidence interval. We didn't. The defensible statement is that mechanically scaling several observed rates produces numbers in that general order of magnitude, while the evidence does not support a precise national count.

After stripping away the category errors, the evidence supports a less dramatic but much sturdier conclusion. Noncitizen voting occurs. Some ballots cast by noncitizens are actually counted, and some cases lead to criminal prosecution. State audits also demonstrate that initial database flags can substantially exceed the number of people ultimately verified as noncitizens, and that some initially flagged voters turn out to be eligible citizens.

Where audits provide usable denominators, detected noncitizen voting has been a tiny fraction of total turnout. That finding does not prove the undetected number is zero, and the decentralized American system does not give us a complete national census of every illegal ballot. It does, however, place a burden on anyone claiming millions of illegal votes: show the records, show the verification method, show the counted ballots and show the denominator.

And if the claim specifically says 'undocumented immigrants,' show the immigration-status evidence too.

The adjective does not get to do the math.

Domenick DeMuro was an elected Judge of Elections in Philadelphia's 39th Ward, 36th Division. This was not a social-media allegation about a blurry security video. DeMuro pleaded guilty in federal court and admitted that a political consultant paid him to add fraudulent votes to voting machines and falsely certify the resulting totals during the 2014, 2015 and 2016 primary elections. DOJ said the added votes increased totals for candidates favored by the consultant and diluted ballots cast by actual voters.^[361]

The consultant was former Democratic U.S. congressman Michael 'Ozzie' Myers. Myers later pleaded guilty himself. According to DOJ, he admitted bribing DeMuro and working with another election official, Marie Beren, in schemes to add fraudulent votes for specific Democratic candidates in elections spanning 2014 through 2018.^[362] Myers would tell election officials which candidates he supported, and fraudulent votes would be added for preferred candidates. He was ultimately sentenced to 30 months in federal prison.^[363]

This is genuine election fraud tied to a Democratic political operative and election officials. It is also important not to inflate what it proves. The convictions establish the scheme and fraudulent votes. They do not automatically establish that every Democratic candidate who received fraudulent votes knew Myers was generating them. 'Benefited from' and 'participated in' are different factual claims, even when social media would very much prefer that they weren't.

The 2018 race in North Carolina's Ninth Congressional District is one of the most useful cases in this article because the misconduct was not merely discovered after an uncontested landslide. Republican Mark Harris led Democrat Dan McCready by only 905 votes after the county canvasses. More than 10,500 absentee-by-mail ballots had been tallied districtwide, and questions about absentee-ballot activity in Bladen and Robeson counties triggered an investigation and a four-day evidentiary hearing.^[364]

The bipartisan State Board of Elections ultimately voted 5-0 to order a new election.^[365] Its formal order concluded that fraud, improprieties and irregularities had tainted the result. Investigators described a coordinated and substantially resourced absentee-ballot operation, and the Board's order said the operation was funded principally through the Mark Harris for Congress Committee.^[364] Critically, the Board did not pretend it could reconstruct an exact fraudulent-vote total. Its order explained that the operation made the precise number of affected votes impossible to determine.^[364] That uncertainty did not rescue the election. Officials refused to certify the original result and ordered voters back to the polls.

So yes: election fraud and election irregularities can be consequential enough to invalidate an election. We have an actual example. No Kraken required.

NC-09 illustrates what evidence looks like when officials conclude that misconduct has genuinely compromised an election. There was a specific contest, a specific absentee-ballot operation, witness testimony, an evidentiary hearing, documentary findings, a 905-vote apparent margin and a formal order from the body responsible for determining whether the result could be certified.^[364][365]

Notice what is missing: a requirement that investigators identify every individual fraudulent ballot before taking the problem seriously. When a coordinated scheme corrupts the process and prevents officials from reliably determining the true result, the legal remedy can account for that uncertainty. The standard is not 'unless you can name exactly 906 fraudulent voters, nothing happened.'

That point cuts in both political directions. Skeptics should not define election fraud so narrowly that even a proven organized scheme becomes irrelevant. Fraud claimants should not define it so loosely that a handful of unrelated illegal votes becomes proof that a statewide result was stolen. NC-09 demonstrates the middle ground: evidence tied to a particular election, evaluated against the actual margin and subjected to a formal adjudicative process.

Now compare those adjudicated cases with the Frio County matter in Texas. In 2025, the Texas Attorney General announced vote-harvesting indictments against nine people, including public officials and political figures.^[366] The defendants named by the state included Manuel Medina, Cecilia Castellano, Raul Carrizales and others. Reporting based on court records identified Medina as a former Bexar County Democratic Party chair and Castellano as a Democratic candidate for the Texas House.^[367]

Those political connections make the case relevant here. They do not turn an indictment into a conviction. In the sources reviewed for this article, these are criminal allegations and must be labeled that way unless later court records establish a disposition. This distinction matters because indictments generate enormous headlines while acquittals, dismissals and plea dispositions can arrive much later. If we freeze the evidentiary status on the day of arrest, we are not tracking fraud. We are tracking press releases.

Put the examples together and a more useful pattern emerges. Philadelphia gives us guilty pleas and a prison sentence in a scheme benefiting Democratic candidates. North Carolina gives us a Republican congressional campaign-linked absentee-ballot operation serious enough that a bipartisan election board unanimously ordered a new election. Texas gives us Democratic-linked defendants facing vote-harvesting charges that, at the evidentiary stage documented here, remain allegations.

Those are three different legal postures and three different kinds of evidence. Flattening them into 'Democrats committed X frauds and Republicans committed Y' would throw away most of what makes the cases informative. The meaningful questions are who did what, what evidence established it, whether ballots were affected, whether the conduct was organized, whether a court or election tribunal adjudicated it, and whether the number or uncertainty was capable of affecting the result. That framework is less useful for partisan point-scoring. It is much more useful for determining whether an election claim is true.

When this article calls a case proven, there should be a guilty plea, conviction, formal adjudicative finding or comparable record behind the word. When it says charged or indicted, the presumption of innocence remains intact. When it says Democratic-linked or Republican-linked, the label describes a documented political connection; it does not silently accuse an entire party or every candidate who may have benefited.

And when someone claims fraud changed an election, we will ask for something resembling the NC-09 evidentiary chain: identify the scheme, connect it to the contest, establish what happened through evidence, compare the problem with the margin, and show what the responsible courts or election authorities concluded. That standard is about to become particularly important, because the next election we examine is 2020.

Six states became the center of most attempts to reverse the 2020 presidential result: Arizona, Georgia, Wisconsin, Nevada, Pennsylvania and Michigan. Biden's certified margins were 10,457 votes in Arizona,^[371] 11,779 in Georgia,^[369] 20,682 in Wisconsin after the recount,^[379] 33,596 in Nevada,^[380] 80,555 in Pennsylvania^[377] and 154,188 in Michigan.^[375]

Those numbers are not proof that every ballot was lawful. They are the scale an outcome-changing fraud claim has to confront. An allegation of ten illegal votes is serious as an enforcement matter, but it cannot erase an 80,555-vote margin. An organized scheme capable of corrupting tens of thousands of ballots would be another matter entirely—but then we would expect evidence commensurate with a scheme of that size: records, participants, affected ballots, audit anomalies, prosecution evidence or findings capable of surviving scrutiny.

That is why the margins belong near the beginning rather than the end. Without them, every irregularity sounds potentially decisive. With them, we can ask the question an election contest actually requires: enough to matter?

Arizona is a particularly useful test because the state did both things political arguments often pretend cannot coexist: it prosecuted actual illegal voters and investigated large-scale fraud allegations without substantiating them. Victor Manuel Aguirre pleaded guilty after illegally registering and voting while his felony voting rights had not been restored; Arizona said his 2020 ballot was counted.^[373] Tracey Kay McKee pleaded guilty after signing and mailing an early-ballot envelope in the name of her mother, who had died before the election.^[374]

Those are not hypothetical cases. Yet the Arizona Attorney General's Office later released records showing that investigators had spent more than 10,000 hours examining 2020 voting irregularities and allegations pushed by prominent election deniers. The investigative summary said the proponents repeatedly failed to provide evidence supporting their claims and that investigated allegations were often speculative, inaccurate or mischaracterized. A small number of matters were submitted for prosecution review, but investigators said the volume aligned with historical trends rather than a widespread conspiracy.^[372]

The certified presidential margin was 10,457 votes.^[371] The evidence therefore supports two simultaneous statements: Arizona had illegal voting in 2020, and the investigations cited here did not establish fraud remotely approaching what would be required to overturn the statewide result.

Georgia's margin—11,779 votes—was the narrowest of these six states, making it the most obvious place to demand unusually strong verification.^[369] Because of the tight presidential race, Georgia's risk-limiting audit became a full manual tally of every ballot. That hand count confirmed the original machine count had identified the correct winner. The Trump campaign then requested a machine recount, which again upheld the outcome, and Georgia recertified the result.^[368]

The state also examined a signature-verification allegation in Cobb County. Investigators reviewed a statistically selected sample of 15,118 absentee-ballot oath envelopes. Ten were escalated for voter interviews; the investigation ultimately reported no fraudulent absentee ballots in the sample, while finding two envelopes that should have been flagged for an opportunity to cure before acceptance.^[370]

That is an important nuance. Election administration was not flawless. The audit found process errors. But 'we found two envelopes that should have gone through the cure procedure' and 'therefore at least 11,780 fraudulent Biden votes existed statewide' are not neighboring conclusions. They live in different statistical solar systems.

Georgia's later political controversy goes beyond ordinary illegal voting and into efforts surrounding certification and alternate electors. We treat that separately in the next tab because somebody casting a duplicate ballot and political actors attempting to alter the certification process are different kinds of conduct.

Wisconsin's post-election fight is also a good place to clean up another common slogan. After the recount, Biden's margin stood at 20,682 votes. Trump challenged the result in Wisconsin court, seeking to exclude categories of ballots in heavily Democratic Dane and Milwaukee counties. The Wisconsin Supreme Court rejected the challenge.^[379]

It is tempting to compress cases like this into 'courts found there was no fraud.' That is too sloppy. Courts decide the claims actually presented under the procedural and substantive rules that govern the case. Some 2020 cases involved standing or jurisdiction. Some involved election-law interpretations. Some claims were rejected for lack of evidentiary support. Some lawsuits did not actually plead a giant voter-fraud conspiracy at all.

So our standard is narrower and stronger: say what the court decided. In Wisconsin, the post-recount legal challenge did not produce a judicial basis for discarding enough ballots to overturn the certified 20,682-vote result. Separately reported voter-fraud referrals should be investigated and prosecuted where warranted, but a referral count is not a conviction count and neither is automatically a 20,682-vote outcome case.

Nevada's certified margin was 33,596 votes. Trump-aligned presidential-elector candidates brought an election contest alleging illegal and improper votes and sought to overturn the result. After an evidentiary proceeding, the district court concluded that the challengers had not proved their claims by clear and convincing evidence, and the Nevada Supreme Court affirmed the denial.^[380]

Again, that does not mean Nevada experienced zero voter fraud. Donald Kirk Hartle later pleaded guilty to voting twice in the 2020 election, including using the ballot issued to his deceased wife.^[381] Mark White was later convicted of attempted voting more than once after trying to vote in Nevada and Arizona.^[382] Those cases are real precisely because investigators developed evidence sufficient to prosecute them.

But two proven or adjudicated examples are not 33,596 votes. The statewide election contest offered a venue for evidence directed at the result itself, and the challengers did not meet the required burden. The existence of Hartle's fraud therefore does not rescue every other Nevada fraud allegation any more than disproving a conspiracy would make Hartle innocent.

Pennsylvania's certified presidential margin was 80,555 votes. State records also rebutted one of the broad numerical claims circulated after the election: Pennsylvania reported 6,945,045 registered voters successfully casting ballots and 6,915,283 votes recorded in the presidential contest, meaning the state did not have more presidential votes than participating voters.^[377]

At the same time, the historical case count did not freeze in November 2020. Matthew Laiss was not convicted until March 2026. A federal jury found that he had cast a Pennsylvania mail ballot on October 31, 2020, after moving to Florida, and then voted again in person in Florida on Election Day. He was sentenced in July 2026.^[378]

That is a useful methodological warning. A post-election report saying only a small number of suspected cases were known at the time is a snapshot, not a prophecy that no future investigation will discover another crime. When later evidence produces a conviction, we add it to the ledger. We do not use the later conviction to pretend the state secretly knew about tens of thousands of additional ballots six years earlier.

The correct update is one proven double voter. Not 80,555 invisible friends.

Michigan's certified margin was 154,188 votes. The state conducted more than 250 post-election audits, and the Bureau of Elections' audit report found no examples of fraud or intentional misconduct by election officials and no evidence that properly programmed and tested tabulation equipment malfunctioned. Those findings addressed major allegations about how Michigan's election had been administered and counted.^[375]

They did not mean no individual person could ever later be proved to have voted illegally. In January 2026, Jose Gargenis Vasquez-Rosa pleaded guilty in federal court. DOJ said the Dominican citizen, who was unlawfully living in the United States, had assumed the identity of a Puerto Rican man, registered under that identity and cast an absentee ballot in Michigan's 2020 election.^[376]

That prosecution is important for two reasons. First, it is direct evidence that an unlawfully present noncitizen cast an illegal 2020 ballot—the kind of case this article refuses to wave away. Second, it demonstrates why an individual illegal vote does not contradict an audit conclusion about tabulation equipment, election-worker misconduct or statewide administration. They are measuring different failure modes.

One illegal voter can coexist with an accurately tabulated election. The denominator does not explode merely because the numerator is no longer zero.

A responsible account of 2020 should retire the line that 'sixty courts found no fraud.' It makes a complicated legal record sound cleaner than it was. Post-election litigation included different plaintiffs, defendants, claims and procedural postures. Some cases failed on standing, jurisdiction, timing or other threshold doctrines. Others involved state-law interpretations. Some reached evidentiary questions. Nevada's election contest is particularly relevant because it did include an evidentiary proceeding and a finding that the challengers had not met their burden.^[380] Wisconsin's high court decided a state-law challenge after the recount.^[379] Other cases elsewhere cannot simply be stapled onto those holdings.

The better question is the one we will use throughout Leiflets: what did this court actually decide? Then ask whether the decision bears on the claim being made. A dismissal for lack of standing is not a factual finding that every ballot was lawful. A full evidentiary rejection is stronger on the merits. A recount is not a criminal investigation. A criminal conviction for one double voter is not a statewide election contest.

The legal record becomes much easier to understand once we stop asking one type of proceeding to prove something it was never designed to decide.

The evidence reviewed here supports a clear answer: we can document illegal votes in the 2020 election, but the audits, recounts, investigations, prosecutions and election-contest records examined for the six decisive states do not establish enough fraudulent votes to reverse any of their certified presidential margins, much less enough to change the Electoral College outcome.

That conclusion is not 'voter fraud never happens.' We just spent several thousand words proving the opposite. It is not 'every 2020 election official performed flawlessly.' Audits documented ordinary errors and procedural issues. And it is not 'every court heard every fraud allegation on the merits.' They did not.

It is a narrower empirical conclusion. Arizona's documented prosecutions do not approach 10,457 votes.^[372] Georgia's full hand tally and requested recount repeatedly confirmed an 11,779-vote outcome.^[368] Wisconsin's post-recount challenge did not overturn 20,682 votes.^[379] Nevada's evidentiary election contest failed against a 33,596-vote margin.^[380] Pennsylvania's later double-voting conviction does not approach 80,555.^[378] Michigan's later noncitizen-voting conviction does not approach 154,188.^[376]

If new evidence establishes additional crimes, add them. Prosecute them. Update the ledger. But an outcome-changing claim has to eventually produce outcome-changing evidence.

So far, the arithmetic remains stubbornly employed.

One of the easiest ways to lose the plot is to describe every Republican elector certificate as though seven photocopies came out of the same printer. They did not. Pennsylvania's certificate says the signers acted on the understanding that they would count only if a final non-appealable court order or another lawful proceeding ultimately recognized them as the duly elected and qualified electors.^[384] New Mexico used similar conditional language, saying it might later be determined that the Republican slate was duly elected and qualified.^[385]

That language does not turn those slates into the official electors—the National Archives still treated the submissions as unofficial—but it materially changes the factual and legal analysis of intent. New Mexico's attorney general later investigated and concluded the state's participants were not subject to prosecution under the law then in effect, while recommending legislation that would more clearly address comparable future conduct.^[386] So the honest summary is not that seven states did the exact same illegal thing. Seven unofficial Republican slates submitted certificates, with important differences in wording and later legal treatment.

Federal prosecutors alleged a coordinated national plan substantially broader than the certificates themselves. The 2023 federal indictment accused Donald Trump and alleged co-conspirators of using knowingly false claims of outcome-determinative election fraud while attempting to organize and use fraudulent elector slates, pressure state officials, enlist the Justice Department and persuade Vice President Mike Pence to alter or delay the January 6 electoral count.^[387]

Those were criminal allegations, not a verdict. The case never reached trial. After Trump won the 2024 presidential election, Special Counsel Jack Smith moved to dismiss the prosecution without prejudice based on the Justice Department's longstanding position that a sitting president cannot be federally prosecuted.^[388] That leaves two bad shortcuts on the cutting-room floor: the indictment cannot be cited as though a jury found every allegation true, and the dismissal cannot accurately be described as a trial acquittal establishing that the alleged conduct never happened. There was no merits trial.

Georgia produced the most sprawling state prosecution. Fulton County's 2023 indictment alleged a broad conspiracy to change the state's election outcome, including pressure on officials and the creation and distribution of electoral documents.^[395] Kenneth Chesebro pleaded guilty in October 2023 to conspiracy to commit filing false documents.^[394]

The prosecution of the remaining defendants later ended without trial. District Attorney Fani Willis was disqualified from the case, a replacement prosecutor later moved to discontinue the prosecution, and in November 2025 Judge Scott McAfee dismissed all remaining charges.^[396] Those two facts coexist. Chesebro has a completed guilty plea. The remaining defendants did not receive trial convictions; their cases were dismissed. Saying everyone was convicted would be false. Saying the entire Georgia matter vanished as though the plea never happened would also be false.

By September 2026, three other state prosecutions had also been dismissed, but the reasons and implications differ. Michigan's district court declined to bind over the remaining 15 defendants after preliminary examination. Attorney General Dana Nessel later declined to appeal and explicitly said the outcome was neither a finding of guilt nor innocence.^[389] The state prosecution is over.

Arizona's existing prosecution was dismissed in 2026, but the dismissal permits refiling and Attorney General Kris Mayes said she intended to return to a grand jury seeking another indictment.^[390] Nevada's August 2026 ruling was different again: Judge Mary Kay Holthus dismissed charges against six Republican electors after finding insufficient evidence from which to infer fraudulent intent and criticizing the state's failure to disclose exculpatory evidence to the grand jury. Nevada's attorney general said he planned an appeal.^[391]

The common word is dismissal. The legal stories underneath it are not common at all.

Wisconsin adds another legal layer. The ten Republican electors settled a civil lawsuit rather than being convicted of a crime. Under the settlement described by Georgetown Law's Institute for Constitutional Advocacy and Protection, they acknowledged that their electoral votes were used as part of an effort to overturn the 2020 result, affirmed that Biden won Wisconsin and accepted restrictions on future elector activity. Chesebro and Trump campaign attorney James Troupis later separately settled related civil claims.^[392]

A different Wisconsin proceeding remains criminal. In June 2026, Troupis, Chesebro and former Trump campaign aide Michael Roman pleaded not guilty to felony forgery charges. Prosecutors allege they deceived the Republican electors about how the certificate would be used; the defendants dispute the allegations.^[393] That case had not produced convictions as of the sources reviewed here. Wisconsin therefore gives us three statuses in one state: civil settlement, acknowledged facts within that settlement, and pending criminal charges against different participants.

On January 2, 2021, Trump spoke by phone with Georgia Secretary of State Brad Raffensperger and other Georgia officials. The call was recorded. During it, Trump said he wanted to 'find 11,780 votes.'^[397] Georgia's certified margin was 11,779 votes, so 11,780 was exactly one more vote than he needed to move from losing to leading the state.^[397]

The arithmetic is striking, but the legal characterization still requires care. A recording proves what was said; it does not by itself resolve criminal intent. Trump repeatedly asserted that he believed he had actually won Georgia and pushed claims that state officials told him were incorrect. Prosecutors later alleged the call formed part of criminal efforts to overturn the result, but no trial produced a verdict on Trump's criminal liability for the call.^[395]

That is why the article should not say Trump committed voter fraud by asking for 11,780 votes. It should say what the evidence establishes: Trump asked Georgia's chief election official to find 11,780 votes after Georgia had certified a 11,779-vote loss and after state officials had repeatedly rejected the fraud claims he raised.

The phone call becomes more informative when placed next to the Georgia evidence from the previous tab. By January 2, the state had already conducted a statewide hand tally and a Trump-requested recount, both of which confirmed Biden's victory.^[368] Georgia officials on the call disputed Trump's allegations and told him his data were wrong.^[397] The request for 11,780 therefore did not occur in an evidentiary vacuum; it came after repeated verification of a result Trump was still contesting.

But precision still beats rhetoric. The strongest factual formulation is not that Trump asked Raffensperger to create 11,780 fake ballots—those are not the literal words in the recording. Nor is it accurate to describe the conversation as merely asking for a routine recount. The record supports describing sustained pressure to identify enough votes to reverse a certified loss while Trump continued asserting fraud claims Georgia officials rejected.

The elector and certification story exposes a blind spot in arguments that define election integrity exclusively as individual voters casting illegal ballots. A presidential election can be attacked or disputed at multiple layers: registration, voting, counting, canvassing, certification, elector appointment and congressional counting. Different conduct implicates different laws and different evidence.

The 2020 record therefore contains two very different realities. We have isolated illegal votes—the kind examined in the previous tab—and organized post-election efforts aimed at certification and the Electoral College process. The latter generated federal and state indictments,^[387] a Georgia guilty plea,^[394] civil settlements,^[392] dismissals on varying grounds^[389][396] and an active Wisconsin prosecution. Those events are important even though they cannot honestly be collapsed into a statistic called fraudulent votes.

If someone says voter fraud, ask whether they mean an illegal ballot. If they mean an alternate-elector certificate, say that. If they mean pressure on a secretary of state, say that. If they mean criminal election interference, point to the charge or adjudication that uses that label. The word fraud is not a magical bucket into which every election controversy can be poured until the distinctions drown.

By the end of this article, a voter-fraud claim should have to survive a fairly simple set of questions. What exactly is being alleged? Is the number registrations, attempted ballots, counted ballots, referrals, charges or convictions? Who investigated it? What did the court or election authority actually decide? How many ballots or electoral votes could the conduct affect? What was the margin? And if the case was dismissed, why was it dismissed?

Those questions do not make fraud disappear. They make real fraud easier to see. DeMuro's bribed election-machine votes are real.^[361] NC-09's absentee-ballot operation was serious enough to force a new election.^[364] Hartle's double voting was real.^[381] Vasquez-Rosa's illegal Michigan ballot was real.^[376] Chesebro's Georgia guilty plea is real.^[394] Michigan's elector charges were dismissed.^[389] Nevada's elector case was dismissed on findings concerning intent and grand-jury disclosure.^[391] Wisconsin's current criminal charges remain allegations.^[393]

That is not a satisfying one-line partisan slogan. It is considerably closer to the record. Voter fraud is real. Election interference is real. Administrative errors are real. False accusations are also real. The job is not to pick the category that helps your side. The job is to identify which one the evidence actually proves.