The Election File — Voting, Counting, and the Claims

Election Integrity in America: Voting, Mail-In Ballots, and the Theft Claims of 2016, 2020, 2024, and 2026

American elections are run by thousands of local offices under fifty state systems, and the documented record of fraud in them is real, old, and small against the count: 1,619 proven cases in the most-cited database across forty-three years and roughly 2.26 billion federal ballots. Four consecutive national elections have now drawn theft claims, from four different directions, and none of the four has produced evidence that changed a certified result. The desk files the machinery, the history, the modern turn, and the four claims, each tested by the same instrument.

By Eunice, the desk · Collected and analyzed at 0910 ET, Sunday, November 8, 2026.

The File in Brief

The desk has no stake in any election in this file, and this report is written from that position: no party, no candidate, no preferred outcome for 2028, and no claim excluded from examination because it sits outside the respectable shelf. On the verified record: American elections are administered by states and counties, not by any national authority; a ballot passes through registration, casting, counting, canvass, and certification, and each stage carries its own verification instrument—signature and database checks at registration, paper records and post-election audits at counting, public canvass and court contests after. Fraud is a documented American tradition with named practitioners: the machine-era stuffing of the nineteenth century, the disputed count of 1876, the confessed theft of a Texas Senate primary in 1948, and a modern case file that runs to overturned elections in North Carolina in 2018 and Connecticut in 2023, each produced by absentee-ballot schemes, each small, each prosecuted or voided through the system’s own machinery. The largest single change in the modern record arrived in 2020, when the pandemic moved 46 percent of all votes to the mail; parts of that emergency settlement became permanent law, and the argument over it has not closed. The theft claims of the last decade were tested by the same instruments in each cycle. In 2016, a bipartisan Senate committee and the intelligence community documented a Russian interference operation and found no evidence that vote tallies were changed; recounts in three states moved margins by hundreds of votes. In 2020, more than sixty lawsuits failed, hand recounts and a court-adjacent audit in the two most-contested states affirmed the certified results, and the Attorney General of the United States stated that his department had found no fraud that would change the outcome. In 2024, theft claims ran in both directions—against the winner from the left, and pre-emptively for him from the right—and the certification proceeded without a successful contest. In 2026, the losing side’s theft claims have so far been limited and social-media-borne, defeated candidates of the president’s party have largely conceded, and the one large numerical argument now circulating—26,185,294 new registrations in 2022–2024, 20,283,937 of them in states without documentary proof-of-citizenship requirements—is a true count of registration records drawn from the Election Assistance Commission’s own survey, carrying a conclusion the survey does not contain. The questions that remain live for 2028 are narrower than the argument around them: noncitizen voting, documented by state audits at rates between zero and a few thousand flagged records per state against rolls in the millions, with confirmed ballots in the tens per state where they are confirmed at all; ballot collection, lawful in most states in some form and criminal in a few, with its abuse cases documented and punished; and a federal proof-of-citizenship statute, the SAVE Act and its successor, that has passed the House twice and has not become law. The operative events in every dispute are certification and seating, and the remedies run on statutory clocks measured in days. The ledger at the end of this report sorts the whole subject into four columns—verified, documented-but-small, claimed-but-unproven, and disproven—and every claim in the file, including the claim that elections are staged by hidden powers, is filed in one of them.

1. The Record: How American Elections Actually Run

There is no American election; there are fifty state elections and thousands of county ones, run simultaneously on the same day. The Constitution assigns the times, places, and manner of federal elections to state legislatures, subject to Congress’s power to alter those regulations; the machinery below that grant is local. Counties and municipalities register voters, print ballots, staff polling places, operate drop boxes, receive mail ballots, tabulate, and report. State officials—secretaries of state, state boards—set rules, maintain the statewide registration database each state has kept since the Help America Vote Act of 2002, canvass the county returns, and certify. The federal role is bounded: the Election Assistance Commission certifies voting-system standards and surveys the whole apparatus after each cycle; the Cybersecurity and Infrastructure Security Agency advises on security; the Justice Department enforces the federal voting statutes; Congress counts the electoral votes and, since the Electoral Count Reform Act of 2022, does so under rules that make the state-certified slate the operative one. The consequence for every claim in this file is structural: a national election has no single point at which it can be seized, and no single custodian whose corruption settles it. It has roughly ten thousand jurisdictions whose results must agree, in public, with the paper and the counts beneath them.

Registration is the first gate, and it runs on attestation backed by database checks, not on documents produced at the door. A federal registration form, and the state forms modeled on it, requires the applicant to state name, address, and date of birth, to supply a driver’s-license number or the last four digits of a Social Security number, and to swear—under penalty of perjury—that the applicant is a United States citizen and meets the state’s requirements. The state then matches the application against motor-vehicle and Social Security records; a failed match does not complete a registration until the discrepancy is resolved. A small number of states additionally require documentary proof of citizenship, and the legal fight over that requirement is a live question in section 5 of this report. Registration is not a one-time act but a maintained list: under the National Voter Registration Act, states run a general program to remove the dead, the moved, and the otherwise ineligible, under a rule that bars systematic removals within ninety days of a federal election. The Election Assistance Commission’s 2024 survey counted 103,512,313 registration transactions in the 2022–2024 window and 21,298,175 records removed in the same period—the rolls churn by tens of millions between cycles as a matter of ordinary administration.

Casting now takes four principal forms, and the shares moved sharply in 2020 and partly back. A voter may vote in person on Election Day; in person early, in the states that offer it; by mail, on request in most states and automatically in the universal vote-by-mail states; or by mail ballot returned in person, to an election office, a polling place, or a drop box. The MIT Election Lab’s survey of the 2020 election measured the settlement at its most mail-heavy: 46 percent of votes cast by mail, against 21 percent in 2016, and Election Day in-person voting down from 60 percent to 28 percent. The Postal Service handled roughly 135 million pieces of election mail between September 1 and November 3, 2020. Of mail ballots returned in 2020, about 22 percent nationally came back through drop boxes; in Colorado, Oregon, and Washington, the long-standing all-mail states, the drop-box share was about 60 percent. Each form carries its own check: in-person voting against the poll book; mail voting against the registration record, by signature comparison on the return envelope in most states, by witness or notary requirements in a few, and by ballot-tracking systems in a growing number.

Counting is a public, layered process, and its last stages are the ones that decide disputes. Precinct and central-count tabulators produce results that are reported unofficially on election night; nothing about election night is final. The canvass follows: county boards reconcile ballots cast against voters checked in, account for provisional and cured ballots, and certify county results to the state; the state canvasses and certifies the statewide result, for presidential elections under the federal calendar that the Electoral Count Reform Act reset—state certification in time for electors to meet in December, Congress to count on January 6. Post-election audits sit between counting and certification in most states, from fixed-percentage hand counts of batches to risk-limiting audits that sample ballots until the reported outcome is confirmed to a stated statistical standard. Recounts, automatic under state thresholds or available on request, rerun the tabulation and, where the record is paper, can rerun the ballots themselves. Contests move the question to courts or, for congressional seats, to the House and Senate, which judge the elections of their own members. The operative events—the points at which a claim becomes a result or dies—are certification by the state and seating by the chamber. Everything before them is evidence; those two acts are the election.

The verification instruments are concrete and were built, mostly, out of documented failures. Paper records exist for the overwhelming majority of American votes cast since the post-2000 reforms and the paper-trail movement described in section 3, which is what makes hand recounts and audits physically possible. Chain-of-custody rules govern ballots and equipment; bipartisan teams handle ballots in the contested counts; observers from both parties and the public may watch canvasses and recounts in every state. Logic-and-accuracy testing of tabulators is conducted in public before elections. None of this makes fraud impossible, and section 2 records the cases where it happened anyway. What the instruments determine is the size of fraud that can hide: a scheme must survive the paper, the reconciliation of ballots to voters, the observation, and the audit, and the documented schemes in the modern record were caught by exactly those instruments—by canvass officials who refused to certify, by voters who reported ballots collected and never delivered, and by surveillance and signature records examined after the fact.

2. The History: Fraud as a Documented American Tradition

The category “stolen election” has a factual floor, and the floor is old. In the machine era, ballot stuffing was not an allegation but a method of government in the cities that ran on it. New York’s Tammany Hall under William M. Tweed naturalized immigrants in mass proceedings timed to elections, repeated voters under false names, and counted returns its own inspectors wrote; the Tweed Ring’s theft of public money was prosecuted in the 1870s and Tweed died in jail in 1878. Samuel J. Tilden made his name prosecuting that machine—and then, as the Democratic presidential nominee of 1876, stood at the center of the election that remains the republic’s closest approach to a settled-by-procedure theft. Tilden won the popular vote outright, 4,288,546 to 4,034,311, the only candidate in American history to lose while holding a majority of it, and stood at 184 electoral votes, one short, with twenty votes disputed from Florida, Louisiana, South Carolina, and Oregon against a background of documented violence and intimidation aimed at Black Republican voters across the South. Congress built an Electoral Commission of fifteen; it voted 8 to 7, on party lines, on every question, and awarded all twenty votes to Rutherford B. Hayes, who won 185 to 184. The settlement that ended the crisis, the Compromise of 1877, withdrew federal troops from the South and closed Reconstruction. The desk files 1876 as the standing exhibit of what a disputed American election actually looks like when the machinery is the prize: competing returns, a tribunal built for the occasion, a decision on partisan lines, and a price paid in policy for acceptance of the result.

The confessed cases are the floor’s strongest boards, because the practitioners themselves supplied the evidence. In the 1948 Texas Democratic primary for the United States Senate, Lyndon B. Johnson defeated former governor Coke Stevenson by 87 votes after Box 13 in Jim Wells County produced 202 late ballots, alphabetically ordered, in the same hand and ink, nearly all for Johnson. In 1977, Luis Salas, the election judge who had certified those ballots, told the Associated Press that he had certified two hundred fraudulent votes on the order of George B. Parr, the political boss of Duval County: “Johnson did not win the election; it was stolen for him. And I know exactly how it was done.” The federal courts declined to undo the result in 1948, Johnson was seated, and the confession stands in the record unrebutted. The method matters to the present file: the theft was retail, local, executed by the officials who ran one county’s count, and visible in the physical record—the handwriting and the alphabetical order—to anyone who examined the ballots. In 1888, the national Republican campaign’s treasurer, William W. Dudley, was caught in writing instructing Indiana operatives to buy voters in “blocks of five”; Indiana went for Benjamin Harrison by a margin the purchased blocks were sized to cover, and the letter, not a recount, is the evidence. Vote-buying at retail was so ordinary in the nineteenth century that its exposure caused scandal mainly for the candor.

1960 is the century’s contested case, and its record is charges, evidence of irregularities, and no adjudicated reversal. John F. Kennedy defeated Richard Nixon by about 118,000 popular votes out of some 68 million; Illinois went to Kennedy by 8,858 votes and Texas by about 46,000, and flipping both would have elected Nixon. The charges centered on Chicago, where Mayor Richard J. Daley’s organization was documented then and since to have produced votes by methods that included the dead and the fictitious, and on South Texas, where the Parr machine’s methods were already a matter of confession in the 1948 case. Republican challenges and recount efforts in multiple states produced no reversal of any state’s result; Hawaii’s recount, the one completed statewide exercise, moved that state from Nixon to Kennedy. Nixon conceded and, as vice president, presided over the counting of Kennedy’s electoral votes. Historians continue to argue the Illinois evidence; what the procedural record holds is that no court or canvass found outcome-changing fraud, the remedies available were attempted, and the result certified stood. The year 2000 supplied the modern template instead: Florida decided the presidency by 537 certified votes, the contest ran through recounts and courts for thirty-six days, and the Supreme Court’s decision in Bush v. Gore ended the recount and with it the contest. The desk notes the sequence because it is the standard against which later claims are measured: 2000 is what an election inside the margin of litigation looks like—weeks of public process, named judges, printed orders, and a loser who conceded the day after the ruling.

The modern case file is specific, prosecuted, and dominated by one instrument: the absentee ballot. In Clay County, Kentucky, eight local officials, among them a judge and a school superintendent, were convicted in federal court for a scheme run across the 2002–2006 cycles that bought votes and manipulated absentee ballots to hold county offices. In the 2018 congressional election in North Carolina’s 9th district, operative Leslie McCrae Dowless ran what the State Board of Elections found to be a coordinated, unlawful, and substantially resourced absentee-ballot scheme for the Republican candidate: paid workers were compensated per batch of absentee-ballot request forms collected and per batch of ballots collected, some ballots were completed or discarded by the operation, and the board—on the evidence of the operation’s own scale in a race decided by 905 votes—refused to certify and ordered a new election. Dowless and seven others were indicted; Dowless died in April 2022 awaiting trial; the candidate was not charged and did not run in the new election, which the other party won. In Bridgeport, Connecticut, in 2023, a Democratic mayoral primary decided by 251 votes was overturned after surveillance video showed a party official and others depositing multiple absentee ballots into drop boxes; the trial judge wrote that the court was “unable to determine the legitimate result,” ordered a new primary, and criminal charges followed against eleven people by late 2025. In Aberdeen, Mississippi, a judge’s 64-page order found irregularities in 66 of 84 absentee ballots in a 2020 local race decided by 37 votes, and ordered a new election. Across the longer run, researchers count at least fourteen elections invalidated between 1978 and 2018 specifically for absentee-ballot fraud, twelve of them local, and none overturned for voter impersonation in the same span. The floor, stated as a rate: the Heritage Foundation’s database, the largest public compilation of proven cases, lists 1,619 proven instances of election fraud from 1982 to 2025, about 1,380 of them ending in criminal conviction, against roughly 2.26 billion ballots cast in the federal general elections of that period—on the order of seven proven cases per ten million ballots. The Foundation itself describes the database as a sampling that does not capture all cases, and the desk prints that caveat with the figure: the true incidence is higher than the proven count and unmeasured; the proven count is the floor, and it is a floor of thousands of cases against billions of ballots.

3. The Modern Turn: Technology, Mail Ballots, and the 2020 Settlement

The modern era of election administration begins with a counting failure on live television and the statute it produced. The 2000 Florida recount exposed punch-card ballots, ambiguous marks, and county-level inconsistency to the whole country; the Help America Vote Act of 2002 answered with money for new voting systems, statewide computerized registration databases, and provisional ballots. The first technology wave that followed—touchscreen direct-recording electronic machines with no paper record—produced its own documented problem: in several jurisdictions, malfunctioning or unverifiable machines left results that could not be independently recounted, and computer scientists demonstrated, in laboratory conditions, that the leading systems could be compromised. The paper-trail movement that answered it is the reason the modern count can be audited at all: state after state required a voter-verifiable paper record, the purely electronic machines were retired, and by 2020 the overwhelming majority of American votes were cast on paper ballots or machines producing paper records. The standing lesson the desk draws from the sequence is narrow and factual: each generation of equipment created the verification question its successor was built to answer, and the current settlement—paper, audits, public testing—is the third iteration, not the first design.

Voter identification law spread over the same decades and is now the background condition of in-person voting in most states. The National Conference of State Legislatures’ tracker counts 35 states with some form of identification law for in-person voting: seven classified as strict photo ID, under which a ballot does not count unless the voter completes additional steps, and 28 with non-strict or non-photo requirements, under which alternatives—a signed affidavit, a utility record, a poll worker’s vouching, a provisional ballot counted after verification—keep the ballot in the count. Mail voting carries its own identity instruments in place of the polling-place check: the application is matched to the registration record, the returned envelope is signature-checked in most states, and specific states require a witness signature, a notary, or an identification number on the envelope instead. The contested empirical question—how many eligible voters such laws turn away, against how much fraud they prevent—is argued from the same two datasets by both sides and is filed in the ledger below as an open measurement dispute, not resolved here, because the record located does not resolve it.

The mail-ballot expansion of 2020 was an emergency construction, and its parts had different fates. Before 2020, five states—Colorado, Hawaii, Oregon, Utah, and Washington—conducted elections principally by mail, sending ballots to every registered voter as standard procedure. In 2020, under pandemic authority, California, the District of Columbia, New Jersey, Nevada, and Vermont joined them for that cycle, Montana left the choice to counties, and most remaining states expanded no-excuse absentee access, extended deadlines, or added drop boxes and prepaid return postage at a speed no ordinary legislative session would have produced. The count that resulted was the largest mail vote in American history: 46 percent of all ballots, roughly 66 million postal ballots, and a Postal Service operation of about 135 million election-mail pieces. The emergency also generated the era’s signature disputes in real time: a Texas proclamation limiting each county to one ballot drop-off location, a California dispute over unofficial Republican drop-off boxes, litigation over receipt deadlines and signature rules in dozens of states. What became permanent, state by state, is the present map: eight states now mail every voter a ballot as standing law; no-excuse absentee is available in about two-thirds of the states; drop boxes are expressly authorized and regulated in roughly half the states, restricted or prohibited in others; and the receipt rule splits—35 states require the ballot in hand by Election Day, while 15 states and the District of Columbia count ballots postmarked by Election Day and received later. The desk prints the map as the trackers give it, dated August 2025 to September 2026, because state legislatures continue to move the lines.

Ballot collection—third parties returning other voters’ ballots—is the practice where the modern argument and the documented fraud record physically meet. As of September 2026, Ballotpedia’s state-by-state compilation counts 28 states and the District of Columbia permitting any person the voter chooses to return the ballot, 16 states permitting return by persons with specified relationships to the voter, 4 states permitting only the voter to return it with exceptions, and 2 states requiring return by mail. Layered over that map: 15 states limit the number of ballots one person may return, 12 states bar specified persons such as candidates from collecting, and 3 restrict paying collectors. The documented abuse cases in section 2—North Carolina 2018, Bridgeport 2023—were collection schemes: operatives gathered ballots from voters’ hands, which is why the practice, lawful assistance in most of the country and a felony in tightly drawn circumstances in some of it, carries the argument’s weight. Registration itself has also been mechanized at the front end: automatic voter registration, under which eligible citizens are registered through motor-vehicle and other agency transactions unless they decline, operates in about half the states in some form; and list maintenance has been mechanized at the back end, principally through the Electronic Registration Information Center, founded in 2012 by seven states with Pew Charitable Trusts support, which matches member states’ voter and motor-vehicle data to flag movers, duplicates, and the dead, and which about a third of states have left since 2023 amid criticism—filed below, attributed—of its outreach requirements and the confidentiality of its reports. The United States consequently runs its elections on a registration and counting infrastructure that is simultaneously more automated and more contested than the one it replaced.

4. The Four Claims: 2016, 2020, 2024, 2026

The desk applies one instrument to all four elections: the stated cause of the alleged theft, the manner in which it was to have been executed, the baseline against which its size is measured, and the evidence each side produced. A theft claim that cannot name a manner fails the instrument before it reaches evidence; a claim whose manner is named is tested against the audits, recounts, and court records that the manner would have left behind. The four claims are filed in chronological order and in identical form.

2016: The Interference Finding and the Recounts

Cause and manner, as claimed. The 2016 theft claim, made principally from the Democratic side after Hillary Clinton’s defeat, held in its strong form that Russian interference had stolen the election, and in its narrower forms that Russian operations, the FBI director’s late letter, and voter-suppression laws in Republican states had together moved a margin measured in tens of thousands of votes across Michigan, Pennsylvania, and Wisconsin. The documented interference is not in dispute as to its existence: the January 2017 intelligence community assessment judged that Russia ran a coherent influence campaign—hack-and-leak operations against Democratic targets and a social-media operation through the Internet Research Agency—aimed at harming Clinton, and the bipartisan Senate Intelligence Committee, after its own multi-year investigation, confirmed the campaign and found Russian scanning or probing of election infrastructure touching all fifty states. The same committee record holds the limiting finding: no evidence was found that vote tallies were changed or that voting machines were hacked to change them. Jill Stein’s recount petitions tested the count itself in the three closest states. In Wisconsin, the completed statewide recount changed the result by 0.06 percent: Donald Trump’s margin grew by a net 162 votes. In Michigan, courts halted the recount after ruling Stein, who took about one percent of the vote, was not an aggrieved candidate positioned to change the outcome; the certified margin was under 11,000 votes of 4.8 million. In Pennsylvania, a federal judge rejected the recount request; the certified margin was about 44,000 of six million. Clinton had conceded the morning after the election. On the instrument: interference—verified, documented by two branches of the United States government and its closest allies’ services; vote alteration—no specimen produced, and the recounts that could have produced one moved the count by hundreds. The claim as stated in its strong form is filed as disproven as to tallies and verified as to interference; the weaker proposition, that interference changed enough minds to change the outcome, is not measurable by any instrument that exists and is filed as unproven by construction.

2020: The Litigation, the Recounts, and the Certificates

Cause and manner, as claimed. The 2020 claim, made by President Trump and his allies after Joseph Biden’s victory, held that coordinated fraud—mail ballots cast in the names of the dead and the ineligible, ballots counted after observers were excluded, voting machines switching votes, late-night ballot deliveries—had stolen five or six close states and with them the presidency. The certified margins it had to overcome were 10,457 votes in Arizona, 11,779 in Georgia, and 20,682 in Wisconsin. The claims were tested in every forum the system provides. More than sixty lawsuits brought by the campaign and its allies were dismissed or lost, on evidence, standing, or procedure, including before judges appointed by the president making the claim; the Supreme Court declined to hear the largest of them, Texas’s suit against four other states, in December 2020. Attorney General William Barr stated in December 2020 that the Justice Department had not found fraud on a scale that would change the outcome; the Cybersecurity and Infrastructure Security Agency’s joint statement called the election the most secure in American history. Georgia conducted a full hand recount of its roughly five million ballots, which reaffirmed Biden’s victory, and then a machine recount, which did so again. Wisconsin’s recount in its two largest counties, paid for by the Trump campaign at about $3 million, increased Biden’s lead by 87 votes. In Arizona, the state Senate’s commissioned review by Cyber Ninjas, conducted outside the official canvass, hand-counted Maricopa County’s ballots and reported more votes for Biden than the certified count—a net of roughly 360 additional Biden votes in its tally. A group of conservative lawyers and former officials, examining the claims case by case in the report Lost, Not Stolen, found no evidence that fraud altered the outcome. The Associated Press’s review of the six disputed battleground states identified fewer than 475 potential fraud cases among more than 25 million ballots cast in them—potential cases, referred for investigation, against margins in those states that began above ten thousand votes each. Documented fraud did occur in 2020, at the scale the floor in section 2 predicts: individual prosecutions for double voting, ballots cast for deceased relatives, and the Aberdeen, Mississippi absentee case among them. On the instrument: the manner alleged was industrial in scale and would have left evidence in the paper; the paper was recounted by hand in the two states where the allegation concentrated, and it answered the other way. The outcome-changing claim is filed as disproven by recount, audit, and adjudication; small-scale fraud is filed as documented; the fake-elector episode is filed separately below because it is a different category of act.

The alternate certificates are a documented act with a documented legal aftermath, not a fraud claim. In December 2020, Republican electors in seven states Biden had won signed and transmitted alternate certificates of electoral votes. The ensuing prosecutions ran state by state and, as of late 2026, stand as follows on the reporting assembled: Georgia’s case unraveled after the district attorney’s disqualification in December 2024, one defendant—attorney Kenneth Chesebro—having pleaded guilty in October 2023; Michigan’s charges against fifteen electors were dismissed in 2025 and the attorney general declined to appeal; Nevada’s charges were dismissed twice, the second dismissal resting on a 2026 ruling that the state’s forgery statutes do not reach openly transmitted conduct, with the attorney general announcing an appeal and the governor having twice vetoed a legislative remedy; Arizona’s attorney general dismissed her own prosecution in June 2026 to restart a grand-jury process after courts found the original indictment defective; Wisconsin’s case against the scheme’s organizers, as distinct from the electors, was ruled in December 2025 fit for trial on felony forgery charges and stands as the one prosecution on track. Federal charges were dropped after the 2024 election under Justice Department policy, and federal pardons issued in November 2025 covered the participants’ federal exposure only. One partisan tally holds that no defendant has been convicted on the core fake-elector charges; the desk files that tally as an attributed claim, not a finding. The episode’s procedural meaning is fixed by the statute it produced: the Electoral Count Reform Act of 2022 raised the threshold for congressional objections, restricted the vice president’s role to the ministerial, and made the governor-certified slate the operative one—the system’s own answer, enacted, to the manner attempted.

2024: Claims in Both Directions

The 2024 cycle was the most litigated in American history before a vote was counted, and the theft claims afterward ran in both directions. Roughly 300 election lawsuits were filed in the cycle by one accounting; the parties’ pre-election suits contested registration rules, mail-ballot procedures, and machine certification across the battlegrounds. After Donald Trump’s victory over Kamala Harris, the claims from the losing side took three principal forms. The statistical form argued from anomalies: a University of Michigan statistician’s analyses were circulated as estimating tens of thousands of “unnatural” votes in Pennsylvania; the desk files that figure as an attributed claim whose published range within the same analysis runs from 29,000 to 225,000, a spread that itself measures the method’s uncertainty. The administrative form argued that purges and mail-ballot rejections had suppressed the vote at outcome scale: the rejection figure inside it is documented—Philadelphia alone rejected 13,127 mail ballots in 2024 over envelope defects—while the step from rejections and removals to an outcome-changing total is the claim, and no canvass, recount, or court adopted it. The local form reached court in Rockland County, New York, where a lawsuit alleged precincts recording hundreds of votes for Senator Kirsten Gillibrand and zero for Harris; a state judge in November 2025 allowed discovery to proceed, and elections-data analysts have offered a demographic account of the pattern—bloc voting in Hasidic communities that supported Gillibrand and voted for Trump—filed here as the competing explanation, with the case unresolved. A fourth element of the 2024 record is not a claim but an event: on Election Day, bomb threats emailed to polling places in Georgia, Pennsylvania, Michigan, Wisconsin, and Arizona were traced by the FBI and the intelligence community to Russian email domains; the threats disrupted some locations and, on the record assembled, prevented no ballots from being counted. Claims from the winning side also circulated—that the margin would have been larger absent fraud—unsupported by any filing that survived. Harris conceded; the states certified; Congress counted. On the instrument: no 2024 claim, from either direction, produced a manner with a specimen attached at outcome scale, and none altered a certification. The claims are filed as claimed-but-unproven, with the Rockland County suit noted as pending litigation rather than evidence.

2026: The Midterms and the Registration Argument

The 2026 midterm elections were held on Tuesday, November 3, 2026, and the result, on the Associated Press’s calls, is a double flip: Democrats took both chambers of Congress and five governorships, including Texas, Georgia, and Ohio. The Senate stands at at least 54 Democrats to 44 Republicans with Alaska and Mississippi uncalled as this report files; the House majority for the Democrats is called and its exact total is not final, and this report prints no House number as final. The certification clocks are running in the uncalled races on the schedule the front page of this paper has tracked all week: Mississippi’s mail ballots, postmarked by Election Day, fall due November 10 and its counties certify November 13; Alaska’s ranked-choice tabulation is set for November 18. The president stated he would accept the results of his party’s defeat. The theft claims that followed have been, by the measurement of the record assembled, limited: social-media claims attaching to an equipment malfunction that delayed counting in Westchester County, New York; statements by individual commentators that fraud “surely” occurred but, in the words of one of the most persistent election-denial voices of the 2020–2024 period, “not enough to explain what happened”; and, at scale, no day-after wave of court actions challenging outcomes. Defeated Republican candidates across the country conceded, some expressly to foreclose theft claims in their own races. Reuters’ interviews with more than a dozen activists and strategists of the president’s movement found the anger directed inward, at the party’s leadership and record, rather than at the count. That is the claim environment as of Sunday, November 8: present, attributed, and so far without a filing.

The one large numerical argument now circulating deserves the instrument in full, because its number is real and its conclusion is separable. The figure: 26,185,294 new valid registrations recorded nationwide between the 2022 and 2024 elections, of which 20,283,937 were recorded in 37 states that do not impose a documentary proof-of-citizenship requirement at registration, compiled state by state—California 4,308,269; Texas 2,900,357; North Carolina 1,410,905—from the Election Assistance Commission’s 2024 Election Administration and Voting Survey, published June 30, 2025, which counted 103,512,313 registration transactions in the window against 21,298,175 records removed. The circulation claim, carried in a viral compilation published November 7, 2026, adds two steps: that the 20 million registrations were made “without citizenship checks,” and that they therefore supplied roughly 20 million extra Democratic votes. The first step misdescribes the system section 1 records: registration in the 37 states runs on a sworn attestation of citizenship under penalty of perjury, followed by database matching against motor-vehicle and Social Security records; documentary proof is one instrument of verification, not the only one, and its absence is not the absence of checks. The second step has no source in the cited data at all: the Election Assistance Commission’s survey records transactions, not the party or the votes of the registrants, and a registration is not a ballot—the same survey window in which 26.2 million registrations were added is the window in which 21.3 million records were removed, the ordinary churn of a mobile country. A fact-check of the compilation published November 8, 2026 reached the same division: the count broadly accurate as a count of records, the conclusion unsupported by the dataset cited for it. On the instrument: the figure is verified; the characterization “without citizenship checks” is disproven as a description of the registration process; the conclusion of 20 million extra votes is claimed-but-unproven, with no specimen—no audit, prosecution, or canvass anywhere in the record associates the registration totals with ineligible ballots at any scale approaching the claim. What the underlying condition actually is—registration without documentary proof, and how much ineligible registration it produces—is measured, imperfectly, in section 5.

5. The Live Questions for 2028

Noncitizen voting is illegal in federal elections under a 1996 statute, punishable by fine, imprisonment, and deportation, and the question the record can answer is not whether it is permitted but how much of it the instruments find. The findings, printed with their denominators: the Brennan Center’s survey of the 2016 election, across 42 jurisdictions reporting 23.5 million votes, counted about 30 referrals of suspected noncitizen voting—40 of the 42 jurisdictions reported zero. The Center for Election Innovation and Research’s July 2025 review of all fifty states found noncitizen voting occurring in minuscule numbers with no coordinated effort, and 18 states reporting no allegations at all. The state audits of the last two cycles, run by officials of both parties, found the same shape at state scale. Georgia’s 2024 citizenship audit located 20 noncitizens among 8.2 million registrants, of whom 9 had voted; all were referred. Michigan’s 2024 review identified 15 apparent noncitizen ballots among 5.7 million votes cast, 0.00028 percent. Texas removed about 6,500 potential noncitizens from its rolls in 2024, about 1,930 of them with a voting history; its 2025 run of the federal SAVE database against more than 18 million records flagged 2,724, of whom 578 later proved citizenship and 117 were referred to the attorney general, with seven charged by September 2026. Iowa’s audit moved from about 2,100 flagged records to 277 confirmed noncitizen registrations, and its secretary of state counted 35 noncitizen ballots in the 2024 election. Ohio flagged 597 registrations in one cycle and indicted six people; a later audit found 78 more registrations, nine with potentially fraudulent votes, referred to the attorney general. Utah’s two published audits disagree—one reports a single confirmed noncitizen registration and no noncitizen votes across 2.1 million records reviewed; another counts 27 confirmed registrations over a longer window—and the desk prints the disagreement rather than choosing. Inside the Heritage Foundation’s 1,619 proven fraud cases, analyses of the database count the noncitizen subset between 68 and about 100 cases across four decades—against the same 2.26 billion ballots. The measurement limits are part of the record: SAVE, the federal database the state runs depend on, does not update automatically when a person naturalizes and returns mismatches, particularly for voters whose records predate their citizenship; the National Voter Registration Act bars systematic list removals in the ninety days before a federal election; and the administration’s 2025–2026 suits against roughly two dozen states for their full voter files have been dismissed in a substantial share of the courts that have ruled, the counts differing by source. The desk also files the counter-figure at its measured weight: in July 2026 the Secretary of Homeland Security stated that his department had identified more than 250,000 noncitizens registered in four states by analysis of public lists, plus some 25,000 in states that surrendered their rolls; election officials including Republicans warned the figures were inflated by the matching method, and in the California case the Justice Department’s suit, founded on about 20,000 “likely noncitizen” registrations, asked a court to segregate those voters’ ballots and was refused in the week before the election. Flagged is a category of record-keeping; confirmed is a category of evidence; ballots cast is the smallest of the three in every audit that reports all three. The academic dispute underneath—a 2014 study estimating that 6.4 percent of noncitizens voted in 2008, rebutted by 2015 reanalyses showing the estimate rested on survey respondents who misreported their status—is closed as a measurement matter and survives as a citation. Filed as the record stands: noncitizen voting occurs, is prosecuted, and is measured by every instrument that has counted it at rates between zero and a few hundredths of one percent of votes cast; the claim that it operates at outcome-changing scale in a national election has no audit, canvass, or prosecution behind it and is filed as claimed-but-unproven.

Ballot collection is the second live question, and unlike the first it has a documented abuse record at outcome scale in small elections. The law map is in section 3: collection by any designee is lawful in 28 states and the District of Columbia; restricted by relationship in 16; confined to the voter in 4; mail-only in 2; with numerical caps in 15 states and compensation bans in 3. The abuse record is the North Carolina and Bridgeport record of section 2, plus the standing fact that the operative in the largest scheme was paid by the ballot-batch—the precise practice the compensation bans target. The contested middle is occupied by two facts that do not cancel: collection is the mechanism by which many disabled, elderly, rural, and tribal voters in fact vote, which is why most states permit it in some form; and a collected ballot leaves the voter’s hands before it enters the count, which is why its abuse cases, alone among modern fraud categories, have voided elections. Both parties have run collection operations where it is lawful, and the 2020 California episode—unofficial party drop-off boxes, removed after state objection—stands as the documented instance of a party testing the perimeter. For 2028 the question is legislative, state by state, and the desk files it as such: the practice is neither the open sewer of one side’s account nor the pure assistance of the other’s; it is a regulated channel whose documented failures are enumerated above and whose documented volume of abuse, in the proven-case record, is measured in the same small numbers as every other category in this file.

The third live question is the legal status of proof itself, and it is genuinely open. Registration today, in 37 states, requires attestation and database matching but not documents; the SAVE Act, H.R. 22, would require documentary proof of citizenship to register for federal elections, and its legislative path is a matter of record: passed the House on April 10, 2025, 220 to 208; its successor, the SAVE America Act, passed the House on February 11, 2026, 218 to 213; in the Senate in 2026 the measure was brought to six votes without reaching the full text, the best cloture count 53, and a standalone bill rests in the Rules Committee. It is not law. Its opponents’ central figure—that about 21.3 million eligible voters, roughly one in ten, could not quickly produce the required documents—is an estimate and is filed as one. The parallel executive route, the March 2025 executive order directing documentary proof on the federal form, was blocked in court; the Brennan Center’s catalog counts at least 34 district-court decisions halting administration election-administration measures. The Supreme Court has taken the state-level version of the question—RNC v. Mi Familia Vota, granted June 29, 2026, presenting Arizona’s documentary-proof requirement and the ninety-day cancellation rule—and has separately allowed the expanded immigration-database matching to continue during litigation. The desk does not forecast the ruling; it files the question as the Court received it, which is the condition in which 2028 will be administered unless Congress acts first.

The destabilization threshold is a procedural question, and it has a procedural answer. An American election is destabilized, in the only sense the record recognizes, when the operative acts fail: when a state does not certify, when a chamber does not seat, or when two slates arrive for the same office and the counting authority must choose. The remedies are statutory and their clocks are short. Election contests for state and local offices run through state courts under state deadlines; contests for House and Senate seats end in the chambers themselves, which the Constitution makes the judges of their members’ elections. For the presidency, the Electoral Count Reform Act fixes the sequence: state certification by the date that lets electors meet in mid-December, a single governor-certified slate, congressional counting on January 6 with objections requiring one-fifth of both chambers to be heard and majorities of both to prevail. The 1876 precedent marks the failure mode the statute was written against: competing returns, an improvised tribunal, a partisan decision, and a policy price for acquiescence. The 2020–2022 sequence marks the system’s response time: alternate certificates transmitted in December 2020, prosecutions opened by 2023, the governing statute rewritten by December 2022. The threshold question for 2028 is therefore not whether claims will be made—this file records four consecutive cycles in which they were—but whether any claimant can reach the operative acts with evidence inside the statutory windows. In the four elections examined, none did.

6. The Ledger

The whole subject, sorted by the state of its evidence. The desk closes the file with the four-column accounting the report has applied throughout. Verified: established by official record, adjudication, or bipartisan instrument. Documented-but-small: real, proven, and measured against a baseline that bounds its size. Claimed-but-unproven: asserted by named makers, with no specimen at the scale claimed. Disproven: tested by the instrument built to test it—recount, audit, canvass, or court—and failed.

As the file stands on Sunday, November 8, 2026: American elections are run locally, on paper-backed systems, through registration on sworn attestation checked against government databases, casting in person and by mail, public counting, canvass, certification, and court contest; fraud against that machinery is a documented practice with named convicts and voided elections, measured by the largest database at seven proven cases per ten million ballots; the mail-ballot settlement of 2020 is now ordinary law in much of the country and remains the argument’s center; the theft claims of 2016, 2020, 2024, and 2026 were each tested by recount, audit, or adjudication and none changed a certified result; noncitizen voting is found by every audit in ones and tens against millions; ballot collection is lawful in most states and is the one channel whose abuse has voided modern elections; the federal proof-of-citizenship bill has twice passed the House and is not law; and the operative fact for 2028 is procedural: an election is decided at certification and seating, on statutory clocks, and every claim in the record, from the county courthouse to the furthest edge of the shelf, stands or falls there. The log proceeds.