Defending Against Bad-Faith Accusations
I. The defensive frame
The case studies that precede this one are dominated by a particular question: did the official reporting accurately reflect what voters did? In most of them, the answer is no — either because an error was missed (Prince William 2020, Monmouth 2022, Butler 2008), or because the existing safeguards were inadequate to surface what went wrong (Baldwin 2002, Warren 2004, Antrim 2020 in its first forty-eight hours), or because the available evidence simply does not let us reach a confident conclusion (Florida 2000, Warren 2004 in its post-lockdown phase). Those cases make the offensive case for Actual Vote: independent verification at the reporting layer is necessary because the existing reporting layer fails in detectable ways often enough that the absence of independent evidence becomes a public problem.
This file is about the defensive case. Even when the official reporting is correct — even when every safeguard works as designed and every official total accurately reflects what voters did — the official record alone is sometimes insufficient to rebut bad-faith accusations of error or fraud. The accusations come from political actors, partisan media, and motivated commentators who have an incentive to delegitimize a result they find inconvenient. They typically point at real procedural anomalies (a software glitch, a late-arriving batch of absentees, a corrected number, an unusual margin) and recast those anomalies as evidence of fraud. The official record can rebut them — and the official record is produced by the same institutions whose accuracy is being attacked, which means its rebuttal is structurally circular to the very audience that most needs persuading.
The methodology already names this problem under Transparency as an independent value and the schismogenesis frame: every existing verification authority in American election life has, for some meaningful share of the public, been captured by one side or the other in the perception of the other side. State boards are dismissed by one side as captured; federal oversight is dismissed by the other. Risk-limiting audits are dismissed by critics who do not trust the auditors. Forensic analyses are dismissed by critics who do not trust the forensics. The verifier and the verified are the same set of institutions, viewed through different lenses.
Actual Vote produces evidence of a different kind. The poll tape is a physical artifact printed by the scanner at the close of polls. A photograph of that tape, captured by a volunteer at the precinct and uploaded to a public corpus, is independent — structurally, not by assertion — of the official reporting chain. A reader who distrusts the Secretary of State can still look at the photograph. A reader who distrusts the federal oversight apparatus can still check whether two numbers match. The evidence does not require trusting an institution. It requires reading a receipt.
This file is illustrative, not exhaustive. Its purpose is to make the defensive use-case concrete with worked examples — drawn from cases the collection already contains, and from additional vignettes selected for analytical strength and political balance. The collection of vignettes is deliberately bidirectional: bad-faith accusations of vote-reporting fraud have come from both directions of the political spectrum in modern American history, and the value of independent precinct-level evidence is structurally identical regardless of which direction the accusation runs. We do not adjudicate any of the underlying disputes. We argue only that, in each case, evidence of the kind Actual Vote produces would have constituted a public, checkable, partisan-independent record against which the dispute could have been resolved on its arithmetic merits.
A note on asymmetry. The historical record is not balanced. The largest, most consequential, and best-documented body of bad-faith vote-reporting fraud accusations in modern American history was produced in the post-2020 cycle, by one political faction, against the Democratic-favoring outcome of that cycle. We do not whitewash this fact. The cases below include both the post-2020 right-aligned accusations that are the most contemporary illustrations of the defensive frame, and the older or smaller left-aligned cases that demonstrate the same structural pattern operating in the other direction. We have not forced symmetry where the historical record does not provide it. We have tried to ensure that the defensive frame is presented as a structural property of independent evidence — not as a critique of any particular faction’s recent behavior.
II. The frame in cases the collection already contains
The defensive frame is not a new addition to the collection. It runs through the existing case studies, often unnoticed beneath the offensive framing. These cases are the historical backbone of the defensive argument; readers should consult the full case studies for the detailed evidentiary record.
Antrim County, Michigan, 2020 — see the case study
The Antrim 2020 error favored Biden by more than 7,000 votes for forty-eight hours before the county published corrections. The corrections were validated by a public hand count that confirmed the corrected totals to within twelve votes; by a GOP-led Michigan Senate Oversight Committee report; by a unanimous Michigan Court of Appeals panel that included two Republican-appointed judges; by a USENIX best-paper academic forensic analysis; and by the Michigan Supreme Court’s denial of further review. None of that prevented the original 4 a.m. screenshot from becoming the central piece of evidence in dozens of post-election lawsuits and in the eventual $787.5M Fox–Dominion defamation settlement. The harm in Antrim was not the original error, which was caught quickly. The harm was the multi-year narrative that grew during the seventeen-day gap between the wrong number and the formally amended record, when no independent evidence existed to anchor the corrected number against the screenshot. AV evidence — captured at every precinct on November 3, 2020 — would have arrived alongside the screenshot rather than seventeen days behind it. See the case study’s Section IX: The Transparency Frame for the full re-frame.
Prince William County, Virginia, 2020 — see the case study
The mirror to Antrim. In Prince William County, Virginia (2020), a ~4,000-vote reporting error — Trump over-reported, Biden shorted — survived certification and a statewide risk-limiting audit that confirmed the counting layer with 99%+ confidence, because the error was in the reporting layer, which that audit did not check; it sat in the official record for more than a year before a new registrar discovered it. Had the discrepancy become public during the unsettled days of November 2020 — a Republican county in a state Biden carried, reporting unexpectedly large Trump numbers — it could have seeded a left-aligned “Republicans stole Virginia” narrative every bit as sticky as the Antrim screenshot. AV evidence at the precinct level would have surfaced the error and provided the same kind of independent precinct-level record regardless of which direction any subsequent narrative tried to grow. The defensive frame does not depend on which side is doing the accusing.
Florida 2000 — see the case study
The Volusia County “negative 16,022 votes for Gore” anomaly was caught only because it was concentrated in a single precinct. The infrastructure to detect a distributed version of the same error did not exist. The election was decided by 537 votes. For two decades after, a substantial body of left-aligned analysis — by Greg Palast, Bev Harris, Mark Crispin Miller, and others — argued, in versions ranging from rigorous to bad-faith, that Florida 2000 was stolen at the reporting layer. Some of those critiques rested on real anomalies; some did not. AV evidence at the precinct level — independent of Katherine Harris’s office, independent of the Bush-Cheney campaign, independent of the FL Supreme Court — would have permitted the post-2000 dispute to be argued from primary-source evidence rather than from the absence of it. The argument might have reached the same conclusion the courts reached, or a different one. The point is that the evidence to argue from would have existed, in citizen hands, outside any institution party to the dispute.
Baldwin County, Alabama, 2002 — see the case study
Don Siegelman won the Alabama governor’s race on election night and gave a victory speech. After the Democratic observer was sent home, officials reopened the courthouse, and by morning ~6,300 votes had vanished from Siegelman’s column at a single precinct, flipping the race to Bob Riley. The recount was blocked by Riley’s certification. The ballots were later destroyed by statute. Democrats have alleged for more than two decades that the election was stolen — sometimes naming Karl Rove, sometimes naming local officials, sometimes naming Diebold. The accusations have hardened into settled left-aligned belief about an Alabama election that the official record cannot definitively resolve in either direction. AV evidence at Bay Minette Precinct 1 — a single citizen photograph of the poll tape at the close of polls — would have been the question-ending piece of evidence. Either the tape totals would match the post-reopening number (refuting the accusation), or they would not (confirming it). Either way, the question would be answerable. Twenty-plus years of speculation rests on the absence of the photograph.
Warren County, Ohio, 2004 — see the case study
The pre-planned lockdown of Warren County’s count, the cited-but-nonexistent terrorist threat, the destruction of records by 56 of Ohio’s 88 counties — these are the case’s narrative center. They sit inside a broader post-2004 left-aligned argument that “Bush stole Ohio,” advanced in book form by Mark Crispin Miller, in House report form by John Conyers, and in magazine form by Robert F. Kennedy Jr. The arguments rested on a combination of real anomalies (the lockdown, the records destruction, exit-poll discrepancies, optical-scan miscounts in specific counties) and contested inferences from those anomalies to fraud. AV evidence at the precinct level across Ohio would have grounded the entire post-2004 dispute in primary-source evidence captured outside any official chain. Whether the analysis ultimately confirmed or refuted any specific claim, the analysis would have been argued from evidence that was structurally independent of the Ohio Secretary of State, the Bush-Cheney campaign, and the Republican-aligned election officials whose conduct was being challenged.
Cook County, Illinois, 2016 — see the case study
The Cook County 2016 audit fraud is itself one of the more disturbing items in the collection — independent monitors watched Board employees erase, replace, and pencil-adjust counts to make them match the certified totals. That is the offensive case. The defensive case sits beside it: the Cook County episode was central to a broader Sanders-aligned narrative that Hillary Clinton had stolen the Democratic primary, in Illinois and in several other states. The narrative was intra-Democratic but consequential — it shaped the 2016 general-election attitudes of a meaningful slice of the American left, and contributed to the legitimacy gap that affected Democratic mobilization that fall. AV evidence at every Cook County primary precinct would have grounded the dispute in checkable arithmetic: the audit fraud at the audit layer would have been visible (as Cook 2016 establishes), but the broader claim that the primary results themselves were fraudulent could have been tested against citizen-held precinct evidence. Some of the broader claim might have been validated. Most of it likely would not have been. Either way, the dispute would have been resolvable.
North Carolina Supreme Court, 2020 — see the case study
The closest statewide race in NC history — 401 votes out of 5.4 million — went through four different counts producing four different margins, all confirming the same winner. Each count was institutional. Each count happened inside the same custody chain that had produced the original number. None of the four counts checked whether precinct totals were faithfully transmitted to the state — that step had 2,662 unchecked opportunities for error. The defensive value of AV in a race this close is structural rather than narrative-bound: had the loser been inclined to allege fraud (in either direction), or had either side’s supporters developed a “the recount was rigged” narrative, the only available rebuttal would have been “the institutional process produced the same answer four times.” AV evidence captured at the precinct level would have been the non-institutional evidence — citizen-held, photographic, distributed — that any future close-election dispute most needs. The 2024 Riggs–Griffin NC Supreme Court contest, which unfolded along similar margins, demonstrates that the underlying dynamic recurs.
III. Right-aligned vignettes — bad-faith accusations against Democratic-favoring outcomes
Wisconsin 2020 — the Milwaukee + Dane County late-night absentee spike
Between 3:26 and 3:44 a.m. on November 4, 2020, the AP election feed showed Joe Biden’s Wisconsin total jump by approximately 150,000 votes (and Donald Trump’s by approximately 32,000) as Milwaukee’s central absentee count and Dane County’s results came in. The spike became the cornerstone of a “ballot dump” narrative that propagated nationally and was amplified for years by Mike Lindell, the Trump campaign, and dozens of social-media influencers. The spike had a mundane technical explanation: Wisconsin law (insisted on by the state’s Republican-controlled legislature) prohibited processing absentee ballots before Election Day; Milwaukee and 38 other communities used a central-count location; Democrats voted absentee at much higher rates than Republicans; the count took until early morning and was reported as a single batch.
Three independent verifiers reached the same conclusion. The nonpartisan Wisconsin Legislative Audit Bureau found no widespread fraud. The Associated Press found no widespread fraud. The conservative Wisconsin Institute for Law and Liberty, a legal organization that has litigated against Democratic causes for years, conducted a ten-month review and explicitly debunked the “spike = fraud” inference. WILL is the most analytically valuable verifier in the case because its institutional incentives run in the opposite direction from its conclusion.
What AV evidence would have contributed: an independent, distributed, citizen-held record of the precinct-level totals from each Milwaukee and Dane County precinct, and of the Milwaukee central-count tape at close of counting. The “vote spike” is a chart-shaped narrative, and chart-shaped narratives do not survive against arithmetic when the arithmetic is publicly verifiable. Any reader could have checked: do the precinct-by-precinct AV tapes match the official precinct totals? Does the central-count tape match the late-arriving batch? The “spike” does not become a different shape of chart, but it becomes a chart that demonstrably reflects what the ballots said, captured in a record nobody had to take WILL’s, the LAB’s, or the AP’s word for.
Sources: Election Integrity Partnership: How real-time vote-count visualizations can mislead; Wisconsin Watch: Was there widespread fraud in 2020?; WPR: Nonpartisan review finds no widespread fraud; Wisconsin Legislative Audit Bureau report; Snopes: Trump-Biden vote dumps debunked.
Maricopa County, Arizona, 2020 — the Cyber Ninjas “audit”
In April 2021, the Arizona Senate’s Republican majority hired Cyber Ninjas — a Florida cybersecurity firm with no prior election-audit experience whose CEO had publicly endorsed “Stop the Steal” claims — to conduct a partisan recount of all 2.1 million Maricopa County 2020 ballots. The process ran until September 2021 and consumed approximately $5.7 million in private and public funds. The Cyber Ninjas final report itself affirmed Biden’s win in Maricopa County and produced 360 more votes for Biden than the official count had recorded. Maricopa County subsequently published a detailed rebuttal documenting nearly 80 false claims in the Cyber Ninjas report. The Arizona Attorney General’s office investigation later concluded the contractors’ material was “speculative in many instances and when investigated by agents and support staff, was found to be inaccurate.” Cyber Ninjas dissolved.
The case is the most ambitious post-hoc partisan verification project in modern American electoral history. It also illustrates the central economic argument for AV: the question the Senate audit was nominally trying to answer — do the official totals match the precinct-level evidence? — could have been answered in days, by any researcher, journalist, or skeptical voter, if AV evidence existed publicly. The seven months and $5.7M were spent on the same comparison that AV produces by default. They were spent because no public independent precinct-level record existed for the comparison to be made against. The Senate could not have found it politically tenable to commission Cyber Ninjas, in the form they did, against a backdrop of public AV evidence already showing the precinct tapes matching the official totals.
The Cyber Ninjas episode is sometimes characterized as a critique of partisan auditing. The defensive AV frame is a different argument. It is not that partisan audits should not exist; an independent audit by interested parties is a reasonable thing for a legislature to commission. It is that the question the audit existed to answer was already answerable from primary-source evidence that AV would have produced for free. The seven-month, $5.7M expenditure is what happens when independent precinct-level evidence does not exist publicly and the political demand for verification is high.
Sources: Wikipedia: 2021 Maricopa County presidential ballot audit; Maricopa County: Correcting the Record (January 2022); AZ Mirror: AG releases withheld 2020 audit findings; NBC News: Cyber Ninjas shutting down; The Century Foundation: The truth behind the Maricopa audit.
IV. Left-aligned vignettes — bad-faith accusations against Republican-favoring outcomes
Stacey Abrams and the 2018 Georgia gubernatorial election
Republican Brian Kemp defeated Democrat Stacey Abrams by approximately 55,000 votes in the 2018 Georgia gubernatorial race. Abrams declined to formally concede; in her ten-day-later statement, she “acknowledged” Kemp’s victory while explicitly refusing to use the word “concede” and characterizing the election as the result of “deliberate and intentional” suppression by Kemp, who had been the sitting Georgia Secretary of State during the election. She referred to the outcome as “stolen” in subsequent public statements, characterized Kemp as the “legal” rather than “legitimate” governor, and used the framing of an illegitimate election in fundraising and political organizing for years.
The bulk of Abrams’s accusations centered on voter suppression — registration purges, the “exact-match” registration policy, polling-place closures and consolidations in predominantly Black counties, and long lines on Election Day. These are pre-tabulation issues that AV does not directly address. But several specific claims did reach into the reporting and counting layers — particularly around the use of paperless DRE machines in counties without auditable trails, allegations of malfunctioning machines in specific precincts, and questions about whether the official totals from particular Black-majority counties could be verified at all. The DRE-machine claims were resolved in subsequent years by Georgia’s transition to paper-trail voting (driven by federal court order in Curling v. Raffensperger). The broader “stolen election” framing has remained a feature of left-aligned discourse about 2018 Georgia.
What AV evidence would have contributed: in counties that used paper-trail equipment in 2018, AV at the precinct level would have produced an independent record of the reporting-layer totals against which the broader “stolen election” framing could be tested. The framing is strongest in its suppression-layer claims, which AV does not address; it is weakest where it has reached into the tabulation and reporting layers, which are exactly what AV does verify. An independent citizen-held corpus would have narrowed the dispute to its actual evidentiary territory — voter access — rather than permitting the looser “the election was stolen” framing to draw rhetorical strength from the absence of independent reporting-layer evidence.
Sources: NPR: Abrams ends her bid for governor; PolitiFact: Stacey Abrams’s claim that Brian Kemp was the architect of voter suppression; Brennan Center: Curling v. Raffensperger; AJC: A timeline of Stacey Abrams and her ‘illegitimate governor’ comments.
New Mexico 2004
George W. Bush won New Mexico over John Kerry by approximately 6,000 votes (49.84% to 49.05%), reversing Al Gore’s 366-vote 2000 win in the same state. The narrow margin and unusual under-vote patterns in low-income and Hispanic precincts in Bernalillo and Sandoval counties — both of which used a mix of DRE and optical-scan voting equipment — fed left-aligned commentary that the result had been manipulated. Bev Harris’s Black Box Voting, Mark Crispin Miller’s Fooled Again, and a contemporaneous body of online analysis argued that the under-vote rate in the presidential race in specific Democratic-leaning precincts was inconsistent with a correctly-functioning tabulation. The state’s investigations were limited; the allegations were never definitively resolved in either direction.
What AV evidence would have contributed: an independent precinct-level record of the reporting-layer totals, captured at the close of polls, against which the under-vote pattern claims could have been argued from primary evidence. The claims rest in part on the gap between expected and actual totals at specific Democratic-leaning precincts; that gap is exactly the kind of thing AV evidence makes checkable. The under-vote rates either were what the precinct tapes said they were (in which case the claim becomes a question about ballot design or voter behavior, not tabulation), or they were not (in which case the claim reaches the reporting layer and is testable directly). Like several of the older left-aligned cases, the dispute persisted in part because the evidence to settle it never existed publicly.
Sources: Wikipedia: 2004 United States presidential election in New Mexico; Wired: New Mexico Sued for Election Fraud; Albuquerque Tribune on under-vote rates (via archive).
Beto O’Rourke v. Ted Cruz, Texas, 2018 (Hart eSlate “vote-flipping”)
In the days before the 2018 Texas Senate election, Texas voters using Hart InterCivic eSlate voting machines reported a screen-rendering issue: when a voter selected straight-ticket Democratic and then attempted to navigate to the next page before the display fully refreshed, the U.S. Senate selection on the summary screen sometimes appeared to show the Republican (Cruz) rather than the Democrat (O’Rourke). The Texas Secretary of State’s office and Hart confirmed the issue was a display-rendering bug that affected the on-screen summary; voters who reviewed the summary screen could correct the apparent flip before submitting, and forensic analysis indicated that the recorded ballot reflected the corrected selection rather than the displayed flip. The issue did not affect the underlying tally.
A brief left-aligned narrative nonetheless characterized the bug as evidence that the machines were “flipping votes from Beto to Cruz.” The narrative had real referent — the bug existed and was visible to voters — and bad-faith amplification, in the form of social-media posts and partisan commentary that conflated a display issue with a tabulation issue. Cruz won the race by approximately 220,000 votes; the bug, even under the most pessimistic plausible accounting of its effect, could not have changed the outcome. What AV evidence would have contributed: an independent record of the precinct-level totals for the Senate race, captured at the close of polls, that voters and analysts could have checked against the alleged flipping pattern. The bug either produced a measurable discrepancy at the reporting layer, or it did not. The question is testable from primary evidence rather than from the absence of it.
Sources: Texas Tribune: Ballot glitches appear to be switching some voters’ Senate selections; PolitiFact: A glitch on Hart eSlate machines caused some Texans to see incorrect votes selected; Texas Secretary of State advisory.
Florida 2018 — Gillum/DeSantis and Nelson/Scott recounts
The 2018 Florida gubernatorial race (Andrew Gillum vs. Ron DeSantis) and U.S. Senate race (Bill Nelson vs. Rick Scott) both went to mandatory recounts under Florida’s 0.5% trigger. Both recounts confirmed Republican wins by narrow margins. The recount process was beset by procedural problems — Broward County’s elections office under Brenda Snipes missed deadlines, mishandled provisional ballots, and discovered ballots in unsecured locations after the fact; Snipes resigned. President Trump and Senator Scott alleged “rampant fraud” from the right; some left-aligned commentary, in parallel, characterized the eventual results as fraudulent or stolen. The Florida Department of Law Enforcement closed its investigation in 2020 finding no evidence of fraudulent intent.
The case is unusual in that bad-faith accusations attached from both directions simultaneously, neither one ultimately validated by official investigation. It is also the case in the collection most directly adjacent to Florida 2000, both in geography and in narrative shape (a contested Florida recount with procedural disorder and multidirectional fraud claims). What AV evidence would have contributed: at the precinct level in Broward, Palm Beach, and other affected counties, an independent record of close-of-polls totals against which both the Republican and the Democratic fraud framings could have been tested. Recount procedure failures are real and documented in the Broward case; they are not the same thing as fraud, and AV evidence provides the discipline to keep the distinction visible.
Sources: PolitiFact: Trump falsely said he stopped 2018 from being stolen; FactCheck.org: Fact-checking Trump’s 2018 election claims; TIME: Florida’s vote-counting controversy spells trouble for 2020; Snopes: Did Broward record more votes than registered voters?.
V. Cases not currently included
A short note on what is not in this file. Several cases meet the analytical bar for inclusion but carry perception costs the collection cannot currently absorb:
Fulton County, Georgia, 2020 → 2026. The State Farm Arena “suitcases of ballots” narrative, the harassment of Ruby Freeman and Shaye Moss, the Giuliani $148M defamation judgment, the OAN settlement, and the January 2026 FBI seizure of Fulton County’s 2020 ballots and tabulator tapes constitute the most prominent contemporary illustration of the defensive frame. The case is deferred from this file because the FBI matter is an active federal proceeding, and writing about that proceeding’s evidentiary basis as “debunked” would place the collection on one side of a live political-legal dispute under a sitting administration. The case will become safer to include after the proceeding resolves. In the interim, the State Farm Arena material and the methodology’s schismogenesis frame stand as adjacent treatments of the same problem.
Wayne County, Michigan, 2020 (TCF Center). The defensive value is real — the “suitcases of ballots” claim, the canvassers’ refusal to certify, the Kraken affidavits, and the broader Detroit-counting narrative were all rejected by Wayne County Circuit Judge Timothy Kenny, all dismissed in court, and all dropped voluntarily by the Trump campaign — but Detroit’s racial subtext makes the case difficult to write without partisan framing risk that exceeds the analytical contribution. Wisconsin 2020 makes substantially the same central-count timing point at lower cost.
The 1960 Cook County election (JFK / Daley). Long-standing right-aligned belief that Daley’s machine stole Illinois for Kennedy. The case is mostly a counting-layer / ballot-handling allegation rather than a reporting-layer one — closer in shape to LBJ 1948 than to any of the cases above. AV does not directly catch counting-layer fraud, so the case is structurally limited as a defensive AV illustration.
Hillary Clinton’s 2016 “rigged election” framing. Clinton has, at various points, characterized the 2016 election as “stolen,” “rigged,” or “illegitimate.” Most of the underlying claims concern Russian interference, James Comey’s October announcement, and voter-suppression activity; few reach the reporting layer that AV verifies. Limited fit for AV’s defensive scope.
The collection will revisit these cases as conditions change.
VI. Why this file matters
The defensive frame is sometimes treated as a softer or less important version of the offensive case for independent verification. We think it is the opposite. The offensive case — AV would have caught this error — depends on errors actually existing. In a healthy election system, errors should be rare. The defensive case — AV would have provided independent evidence to anchor the post-election dispute — does not depend on errors existing. It depends only on the existence of disputes, and disputes about electoral legitimacy are now a recurring feature of American political life regardless of whether errors occur.
The post-2020 cycle made this asymmetry visible at scale. The actual reporting-layer errors of the 2020 election, considered as a body, were small and quickly corrected. The political dispute that grew on top of them was enormous and is still ongoing. The harm caused was disproportionate to the underlying technical phenomenon by orders of magnitude. That asymmetry is what the offensive frame, taken alone, cannot explain. The offensive frame says: AV catches errors. Yes — but the harm in 2020 was not caused by uncaught errors. The harm was caused by the gap between corrected official records and the independent evidence that would have made the corrections evidentiarily authoritative to a public that had no reason to take the corrections on institutional faith.
The cases in this file — those drawn from the existing collection, those added as vignettes — are the worked examples. Each one is a moment in American electoral history when an official record was correct, or at least never disproven, and when bad-faith accusations of error or fraud nonetheless attached to it and persisted. In each case, independent precinct-level evidence captured in citizen hands at the close of polls would have changed the trajectory of the dispute. Sometimes — as in the strongest cases — the evidence would have foreclosed the bad-faith narrative entirely, making it difficult to maintain in the face of publicly checkable arithmetic. Sometimes the evidence would have narrowed the dispute to its actual evidentiary territory rather than permitting it to draw rhetorical strength from the absence of independent records. Sometimes — as in the hardest cases, where the evidence might have confirmed the suspicion rather than refuted it — the resolution would have required acknowledging that the suspicion was correct. Either way, the dispute would have been resolvable.
The defensive value is bidirectional and structural. It does not depend on which side of the political spectrum the accusation comes from. It does not depend on whether the accusation is correct. It depends only on the existence of independent primary-source evidence captured outside the institutions whose accuracy is being challenged. That is what Actual Vote produces. It is what no other available verification mechanism produces at the precinct level, in real time, in a form ordinary citizens can verify themselves. The cost of not producing it is the cost displayed in this file, multiplied across cases the file does not name and across cases that have not happened yet.