Case Studies
America Counts maintains this case study collection to ground its work in real evidence. Each case examines a documented situation involving vote-reporting error, fraud, or vulnerability, and asks: what happened, what do we know, what don’t we know, and how would independent verification by Actual Vote have changed the picture.
The cases are not meant to make any case for any party. They are meant to make the case for independent verification — the kind that Actual Vote provides — by showing concretely what happens when it is absent. The full framework for how we select and present them is in our methodology page.
Independent poll-tape verification would have surfaced every case below — in real time, without waiting for an official to notice. Most were equipment or process error rather than deliberate fraud; a few are structural vulnerabilities; two remain unresolved because the evidence was destroyed. Expand any row for detail.
-
1860s Tammany Hall fraud
ScaleCity-wide machine fraud; ~65,000 rushed naturalizations before the 1868 election
At stakeControl of New York City government
What went wrongReturning boards "announced the result in bulk" without counting ballots; organized repeat voting at scale.
How it resolvedSustained for years until the Tweed machine finally collapsed under prosecution.
How AV would have helpedAn independent, public record of each precinct’s count would have made "announcing the result in bulk" impossible to do unchallenged.
-
1948 LBJ Senate Race fraud
Scale87-vote statewide margin; 202 fabricated votes from one precinct (Box 13)
At stakeA U.S. Senate seat — and, downstream, the presidency
What went wrongReturns "amended" six days later: 200 new votes for Johnson, 2 for Stevenson, in one hand, in different ink.
How it resolvedStood. Johnson was seated and rose to the presidency; the election judge confessed decades later.
How AV would have helpedThe precinct’s own election-night tally, captured independently, wouldn’t match the return amended six days later — the 200 added votes would have nowhere to hide.
-
1998 Gaston County error
ScaleA third of precincts reported zero; a software bug capped totals at 32,000
At stakeAccurate county results — and public trust in them
What went wrongFaulty data cartridges and a 32,000-vote software cap garbled precinct reporting.
How it resolvedThe board chair opened the entire recovery to public observation; resolved without protest. The positive case.
How AV would have helpedIndependent precinct tapes would have given a correct reference to rebuild from while the cartridges were diagnosed — and confirmed the recovered totals.
-
2000 Florida error
Scale537-vote statewide margin; one Volusia precinct (585 voters) reported −16,022 for Gore
At stakeThe presidency
What went wrongA faulty memory card made a single precinct transmit a negative 16,022-vote total to the county.
How it resolvedCorrected, but not before it moved the network calls and Gore’s first concession on election night.
How AV would have helpedA precinct-level tape-vs-report check would have flagged the impossible −16,022 instantly — and caught the smaller, distributed errors no statewide total could reveal.
-
2002 Baldwin County unresolved
Scale~6,300 votes vanished from one column at a single precinct — enough to flip the race
At stakeThe Alabama governorship
What went wrongTotals at one precinct changed overnight after the opposing party’s observer was sent home.
How it resolvedRecount blocked; ballots later destroyed by law. What actually happened can never be answered.
How AV would have helpedAn independent election-night tape from that precinct would be the one record that survived — turning an unanswerable dispute into a checkable comparison.
-
2004 Warren County unresolved
ScaleCount conducted under lockdown; 56 of 88 Ohio counties later destroyed records
At stakeVerifiable presidential results in a decisive swing state
What went wrongOfficials barred all observers citing a terror threat the FBI knew nothing about — planned a week ahead.
How it resolvedRecords destroyed despite a federal retention law; the central questions are permanently unresolvable.
How AV would have helpedTapes photographed at the precincts would exist outside the county’s control — independent evidence no building lockdown or record purge could erase.
-
2008 Butler County error
ScaleVotes silently dropped on upload; the bug class spanned ~a decade across 34 states
At stakeReliable counts anywhere GEMS was deployed
What went wrongA GEMS logic error dropped votes during memory-card uploads — every upload still marked "successful."
How it resolvedIndependent investigation proved the upload bug; the vendor later admitted it had persisted for roughly a decade.
How AV would have helpedComparing precinct tapes to the uploaded totals would have exposed the silent drop on election night, instead of years later.
-
2013 Clarkson v. Lehman vulnerability
ScaleFive years and two lawsuits; the underlying tapes never opened to inspection
At stakeThe public’s right to inspect election records
What went wrongA statistician’s request to inspect the paper trail was denied; courts held the records were not open to inspection.
How it resolvedEvery court ruled against access; the tapes were destroyed before they could be examined.
How AV would have helpedCitizen-captured tapes are public by design — the evidence the courts wouldn’t release would already be in the open, independent of the county’s custody.
-
2015 Shelby County error
ScaleA tape showed 546 votes; official said 330 — ~1,001 votes dropped across precincts
At stakeAccurate local results, concentrated in majority-Black precincts
What went wrongA reporting/aggregation bug dropped votes between the poll tape and the official totals.
How it resolvedOne citizen’s photo exposed a systemic drop; the elections administrator retired under pressure. The closest real-world AV demonstration.
How AV would have helpedThis is the AV method itself: a photographed tape (546) set against the official total (330) is exactly the comparison that exposed the drop.
-
2016 Cook County vulnerability
ScaleAudit hand counts altered to match official results, on the record
At stakeThe integrity of the post-election audit itself
What went wrongAudit staff erased, replaced, and pencil-adjusted hand counts to match official totals; monitors documented it.
How it resolvedThe Seventh Circuit ruled the audit could not change the certified result no matter how it was conducted.
How AV would have helpedIndependent tape evidence is an external check the in-room audit can’t quietly "adjust" to match — the comparison doesn’t depend on the people running it.
-
2016 Fraction Magic vulnerability
ScaleDemonstrated on real GEMS software; can shave any chosen percentage of a candidate’s votes
At stakeAny race tabulated on the affected reporting system
What went wrongGEMS stores per-candidate totals as decimals; an attacker shows the rounded whole on screen but reports a fractionally reduced total.
How it resolvedA demonstrated vulnerability, not a known deployed attack; an independent poll-tape comparison would expose it immediately.
How AV would have helpedThe printed poll tape shows whole numbers; comparing it to the GEMS-reported total would reveal any fractional shaving at once.
-
2020 Prince William County error
Scale~4,000 phantom votes; the statewide RLA reported >99% and missed it
At stakeCorrect presidential and down-ballot totals
What went wrongA formatting incompatibility between scanner tapes and the state reporting system added ~4,000 votes.
How it resolvedFound over a year later by a new registrar via a records request. The textbook reporting-layer case.
How AV would have helpedA tape-vs-report comparison would have flagged the ~4,000-vote discrepancy on election night — not fourteen months later.
-
2020 Antrim County error
Scale~22,000-voter county; first posted Biden 7,769–4,509, corrected to Trump +~3,800
At stakePublic confidence — the 4 a.m. screenshot seeded national misinformation
What went wrongThe county EMS aggregated scanner memory cards using an updated ballot definition the scanners never received.
How it resolvedCorrected within 48 hours and confirmed by a hand count to within 12 votes — but the screenshot had already traveled the world.
How AV would have helpedAn independent precinct tally posted alongside the county’s 4 a.m. number would have shown the mismatch in real time, before the screenshot could spread.
-
2020 Georgia error
Scale~5,900 votes missing across four counties; statewide margin of 11,779
At stakeThe presidency (the state’s 16 electoral votes)
What went wrongPrecinct memory cards were never uploaded — tabulated correctly, but absent from the official totals.
How it resolvedSurfaced only because the margin triggered a full statewide hand recount.
How AV would have helpedComparing captured tapes to the reported totals would have surfaced the ~5,900 missing votes immediately, without waiting for a margin-triggered recount.
-
2020 NC Supreme Court vulnerability
Scale401 votes out of 5.4 million; four counts produced four margins
At stakeA state supreme court seat
What went wrongNone of the four counts verified that precinct totals were faithfully transmitted to the state.
How it resolvedAll four counts confirmed the same winner; the reporting-chain question was never directly checked.
How AV would have helpedAV supplies the one check none of the four counts performed — verifying precinct tapes against the transmitted state totals.
-
2022 Monmouth County error
ScaleSix flash drives counted twice; the wrong winner was seated for two months
At stakeA local elected office
What went wrongA removed anti-double-counting patch let six flash drives load twice on election night.
How it resolvedCorrected after ~two months; the official audit had reported 100% accuracy.
How AV would have helpedPrecinct tapes set against the reported totals would have caught the doubled drives on night one, before the wrong winner was seated.
-
2022 DeKalb County error
ScaleA candidate shown 0 votes in her home precinct; error repeated across ~a dozen precincts
At stakeA county commission seat
What went wrongPrecinct results misreported a candidate’s real votes as zero — a counting-layer error severe enough to be visible on the tape itself.
How it resolvedA Memorial Day hand count showed she had actually won outright with 43%; the result was corrected. A voter’s poll-tape photo surfaced it.
How AV would have helpedThe public tape content itself shows the candidate’s real votes — exactly what flagged the zero-vote error here.
-
2024 Gillespie County error
ScaleA hand-counted primary that ran to 4 a.m.; reconciliation errors and duplicate entries
At stakeAccurate primary results
What went wrongEven with hand counting, the transcription chain produced reconciliation errors and duplicate precinct entries.
How it resolvedTook weeks to resolve — showing the reporting layer is fragile regardless of how the counting is done.
How AV would have helpedIndependent tapes give a fixed reference to reconcile against, catching the transcription and duplicate-entry errors that took weeks to untangle.
Here’s a map of where each case study comes from. Problems with vote reporting have been discovered in many states over more than a century. Hover or select a marker to open its case study.
A note on what these cases are for
Most of these case studies involve specific elections, specific parties, and specific candidates. We selected them based on how well they illustrate distinct kinds of vote-reporting vulnerability — not based on which party they portray favorably or unfavorably. Across the collection, no political side is spared and no political side is singled out. The point of any individual story is never “this party cheated.” It is “here is a situation where independent evidence from Actual Vote would have given everyone, regardless of party, a clearer picture of what happened.”
For a deeper discussion of how America Counts maintains its non-partisan commitment in this work, see our Non-Partisan Election Advocacy page.