Case 14 — Georgia, 2020

14.1 — Introduction

On November 3, 2020, approximately five million Georgians cast ballots in the presidential election. Joe Biden won the state by 11,779 votes — a margin of 0.24% — flipping Georgia blue for the first time since Bill Clinton in 1992. What followed was the most intensive post-election scrutiny of any state in modern American history: a full hand tally of every ballot, a machine recount, and a signature audit.

All three processes confirmed the same winner. But they also revealed something that received far less attention: thousands of valid votes had never made it into the official count. In four counties, memory cards containing vote totals from precinct scanners failed to upload into the county reporting systems. The votes were tabulated correctly at the precinct level — the scanners counted them, and printed the totals on their tabulator tapes — but those totals were never transmitted to the county election offices. The votes simply vanished from the reporting chain.

These were not allegations. They were confirmed, documented reporting-layer failures, discovered only because a statewide hand recount happened to take place. Without it, those votes would still be missing.

This case study examines several vote-reporting incidents from the Georgia 2020 election: the memory card failures in Floyd, Fayette, Douglas, and Walton counties; the unsigned and missing tabulator tapes in Fulton County; and the broader procedural inconsistencies documented by the Carter Center across 25 counties. Together, they illustrate why independent, systematic auditing of vote reporting matters — and why selective auditing driven by partisan suspicion is no substitute.

14.2 — Background: How Georgia Reports Votes

Georgia’s voting system in 2020 was new. In 2019, the state replaced its paperless Diebold Accuvote TSX touchscreen machines — which had been in use since 2002 and produced no paper trail — with Dominion Voting Systems ballot marking devices (BMDs). Under the new system, a voter uses a BMD to make selections on a touchscreen, and the device prints a paper ballot containing a plain-text summary of the voter’s choices along with a QR code. The voter reviews the paper ballot and feeds it into a scanner/tabulator, which reads the QR code and records the vote. The paper ballot is retained for audits.

The scanner/tabulator is the critical piece of hardware for understanding the reporting process. Each scanner maintains a running count of votes as ballots are fed into it throughout election day. When the polls close, the scanner prints a tabulator tape — a receipt-like record showing the total votes for each candidate on that machine. Under Georgia Rule 183-1-12-.12, a poll manager and two witnesses are required to print three copies of this tape and sign each one, certifying it as a true and correct record.

The vote totals are also stored on a removable memory card inside the scanner. After the polls close, election workers transport these memory cards to the county election office, where the data is uploaded into the county’s election management system. The county aggregates the results from all its precincts and reports them to the Secretary of State.

This two-step process — physical printout at the precinct, electronic transmission to the county — creates both a safeguard and a vulnerability. The tabulator tape is an independent record of what the scanner counted, generated before anyone has an opportunity to alter the data. But the electronic transmission depends on memory cards being properly handled, transported, and uploaded. If a memory card fails to upload, the votes it contains are missing from the official results — even though the tabulator tape at the precinct recorded them correctly.

14.3 — The Election and Its Aftermath

Biden’s initial lead after the first machine count was approximately 14,000 votes. Georgia immediately became the epicenter of national controversy. Allegations of fraud and irregularities, focused overwhelmingly on Fulton County (which includes most of Atlanta), began circulating within hours of the polls closing.

In September 2020, Georgia’s State Election Board had adopted a new rule requiring a risk-limiting audit (RLA) following each November general election. The Secretary of State selected the presidential race for the inaugural audit. Under the RLA framework, a statistical sample of ballots would be hand-counted and compared against the machine totals, with the sample size determined by the margin of victory and a specified risk limit.

But with Biden’s margin at just three-tenths of one percent, the mathematics of the RLA pointed toward sampling approximately 1.5 million of the state’s roughly five million ballots — a logistically daunting prospect. On November 11, Secretary of State Brad Raffensperger announced that the audit would instead take the form of a full hand tally of every ballot: essentially an RLA with a risk limit of zero. This would be the largest hand count of an election race in United States history.

Counties had less than 48 hours to prepare. The hand tally began on November 13 and was completed by November 18, with certification on November 20. The result confirmed Biden’s victory, with net adjustments of approximately 1,200 votes in Trump’s favor, narrowing the margin to 12,284.

Subsequently, the Trump campaign requested a machine recount under Georgia law (permitted when the margin is less than 0.5%). This recount, conducted November 21 through December 7, produced the final certified margin: Biden by 11,779 votes. A third process — a signature audit of absentee ballot envelopes — was conducted in Cobb County as a pilot. The Georgia Bureau of Investigation found only two mismatched signatures out of all examined.

Three separate processes. The same winner each time. But the hand tally revealed something that neither of the other processes would have caught on its own.

14.4 — The Missing Votes: Memory Card Failures

Late in the hand tally process, four counties discovered that previously uncounted ballots were sitting in their precincts. In each case, the problem was the same: vote data stored on memory cards from precinct scanners had not been uploaded into the county election management system. The scanners had counted the ballots and printed the totals on their tabulator tapes, but the electronic link between the precinct and the county broke down. The Carter Center, which deployed 52 independent monitors to 25 counties during the hand tally, documented these discoveries in its final report.

Floyd County: 2,600 uncounted ballots. During early voting on October 24, a scanner at the Floyd County Administrative Building in downtown Rome jammed. Election workers replaced it with another scanner and moved approximately 5,000 ballots from the jammed machine to the replacement. But about 2,600 of those ballots were never re-scanned. The memory cards from the original scanner did not transfer properly into the election software.

Gabriel Sterling, a senior official in the Secretary of State’s office, said the failure was “not an equipment issue but a person not executing their job properly.” The Secretary of State asked Floyd County’s Chief Clerk of Elections, Robert Brady, to resign. Brady was in COVID-19 quarantine and unavailable for comment at the time. Floyd County subsequently rescanned all 8,100 early voting ballots from that location, confirming that the hand-count totals matched the rescan. The net effect of the discovered ballots was less than 800 additional votes for Trump.

Fayette County: 2,755 uncounted ballots. A memory card from a scanner was never uploaded to the county system. The card contained 2,755 votes: 1,577 for Trump, 1,128 for Biden, 43 for Jorgensen, and 7 write-in votes.

Douglas County: 293 uncounted ballots. A memory card from an Election Day precinct was not uploaded, leaving 293 votes out of the initial reported results.

Walton County: 284 uncounted ballots. A memory card was discovered containing 284 votes that had not been included in the initial results.

In total, nearly 6,000 votes across four counties were missing from the official results due to memory card upload failures. Three of the four counties reconciled their counts against the number of voters who had voted, recounted the affected ballots, and certified corrected results in advance of the state certification on November 20. The Carter Center report noted that the missing ballot problems “appear to be cases of reconciliation procedures being inadequate to track the range of voter options for submitting ballots.”

The critical point is this: every one of these errors was a reporting-layer failure. The scanners tabulated correctly. The tabulator tapes printed correctly. The paper ballots existed and were intact. The only thing that went wrong was the transmission of the electronic results from the precinct to the county. And the only reason the errors were caught was that Georgia happened to conduct a full hand recount of every ballot — an extraordinary, expensive process that no one expected and that was triggered solely by the tightness of the margin.

The same class of failure occurred in Prince William County, Virginia that same year (see Case 12 — Prince William County, Virginia, 2020). There, scanners counted correctly and printed correct tapes, but a formatting incompatibility between the scanners and the state reporting system introduced approximately 4,000 phantom votes. That error went undetected for over a year — because Virginia, unlike Georgia, had no hand tally to catch it.

14.5 — The Unsigned Tapes: Fulton County

In December 2025 — five years after the election — a different kind of reporting-layer failure came to light in Fulton County, Georgia’s most populous county and the one that had been the focus of the most intense scrutiny.

Fulton County admitted before the State Election Board that more than 130 tabulator tapes from the 2020 election were unsigned. These tapes covered virtually all of the county’s early in-person votes — more than 315,000 ballots. Under Georgia Rule 183-1-12-.12, a poll manager and two witnesses are required to sign each tabulator tape, certifying it as a true and correct record of the vote totals produced by that scanner.

That was not done.

Additionally, 10 of the county’s 148 total tabulator tapes are missing entirely, covering more than 20,000 votes. These became the subject of an earlier, separate complaint.

At the December 9, 2025 State Election Board hearing, Fulton County attorney Ann Brumbaugh did not dispute the violation. She characterized it as an administrative error, noting that the county has since implemented enhanced training, new standard operating procedures, and a self-initiated investigation process for unsigned tapes. Secretary of State Raffensperger stated that the unsigned tapes did not invalidate the underlying votes, noting that all voters had been verified with photo ID and lawfully cast their ballots.

Others viewed the situation more severely. David Cross, the complainant, called the unsigned tapes “catastrophic breaks in chain of custody and certification.” State Election Board member Janelle King said that “at best this is sloppy and it’s lazy, at worst it could be egregious.” The Board voted 3-0 to refer the case to the Georgia Attorney General’s Office for possible sanctions, requesting a $5,000 fine for each unsigned tape — potentially totaling$ 670,000 or more.

The significance for this case study is not the question of whether the unsigned tapes indicate fraud — every investigation of the 2020 Fulton County results has concluded they do not. The significance is what unsigned and missing tapes mean for the evidentiary record. A tabulator tape is the closest thing to a definitive precinct-level record of what the scanners actually counted. When that record is unsigned, there is no certified chain of custody connecting the tape to the count. When the tape is missing entirely, the record simply does not exist. In either case, anyone wishing to verify the reported results at the precinct level is left without the primary piece of evidence.

14.6 — What the Carter Center Observed

The Carter Center’s monitoring of the hand tally provides the most detailed independent account of how the process actually worked in practice. The Center deployed 52 trained monitors to 25 of Georgia’s 159 counties, covering 60.27% of the state’s total votes. Monitors submitted 509 audit board reports, 96 vote review panel reports, and 54 general observation forms over five days of observation.

Their overall assessment was positive: 86% of daily general observation forms rated the process as well-managed and 81% described the check-in/out process as well organized. County election officials adapted impressively to the challenge of organizing the largest hand tally in American history with less than 48 hours of preparation, during a pandemic.

But the monitors also documented widespread procedural inconsistency. In 18 of the 23 counties where audit board activity was observed, counting practices varied from those demonstrated in the Secretary of State’s training video. Some audit boards had one member sort and the other count; others sorted and counted independently. Some counted in groups of 10, others in groups of 20 or 25. In some counties, one member sorted and called out candidate names while the other watched; in one county, the two team members counted independently without looking at each other’s work, prompting the election superintendent to make them start over.

Data entry procedures also varied widely. The official protocol called for two-person teams, with one reading aloud from the batch sheet and the other confirming the entry into the computer. In practice, many counties used single-person data entry. Gwinnett County expanded from one or two laptops to 15 two-person data entry teams as the scope of the task became clear.

Ballot security presented challenges. In some counties, ballot containers were unsealed by check-in staff and distributed among multiple audit boards rather than remaining sealed until the receiving audit board opened them. In one county, the Carter Center monitor reported that check-in staff were absent from their stations for 15 to 20 minutes at a time, during which ballot boxes were left unguarded. In one metro county, open boxes of ballots were left under audit board tables during lunch breaks.

The Carter Center also documented the behavior of political party monitors, who had the same access as Carter Center monitors to observe the hand tally. In nine counties, Carter Center monitors noted examples of Republican monitor interference or hostility: leaning over audit boards, using cellphones on the audit floor (prohibited by rules), photographing ballots, hovering over data entry operators, and in one county, accusing two African American audit board members of fraud. In four counties, disruptions by monitors or observers resulted in law enforcement being called. In one county, an audit board member was reduced to tears by persistent criticism from a Republican monitor.

Carter Center monitors also reported that many party monitors — from both parties — did not appear to understand the RLA process. Comments from Republican monitors included statements such as “This is just a PR stunt” and complaints that the audit was “just looking at marks on a paper” rather than checking signatures or voter eligibility.

None of these procedural variations changed the outcome. The Carter Center concluded that ballot integrity was maintained and that county officials did an admirable job under extraordinary circumstances. But the observations illustrate an important point: the hand tally was an improvised process, conducted at unprecedented scale under extreme time pressure, with inconsistent procedures across counties. It worked — but it was not the kind of systematic, standardized audit that could be routinely replicated.

14.7 — The Asymmetry Problem

The three post-election processes in Georgia — the hand tally, the machine recount, and the signature audit — were triggered by the closeness of the margin and the intensity of political demand, primarily from Republicans alleging fraud in Democratic-leaning counties. The allegations focused overwhelmingly on Fulton County (Atlanta) and, to a lesser extent, DeKalb County — both heavily Democratic and majority-Black.

The irony is that the actual reporting failures discovered during the hand tally were concentrated in different places. Floyd County, where 2,600 ballots went unreported, voted for Trump by a roughly 3-to-1 margin. Fayette County, where 2,755 ballots went unreported, also leaned Republican. These errors were discovered not because anyone suspected problems in Republican counties, but because the hand tally happened to cover every county in the state.

Meanwhile, 134 of Georgia’s 159 counties had no Carter Center monitors present at all. The counties in Georgia’s Black Belt — the band of rural, majority-Black counties stretching across the south-central part of the state — received no particular scrutiny from any of the post-election processes. This is not to suggest that reporting errors existed in those counties. It is to observe that we have no way of knowing, because nobody looked.

This asymmetry — intense scrutiny where political suspicion points, and no scrutiny where it doesn’t — is a structural feature of how American elections are typically audited. Post-election challenges are initiated by candidates and parties, who focus their resources on the jurisdictions they believe will yield the most favorable results. The places that don’t attract political attention go unexamined.

An Actual Vote analysis does not depend on which party demands it or which counties attract suspicion. It generates evidence from every precinct where volunteers participate, regardless of partisan considerations. Had AV been deployed across Georgia in 2020, it would have produced evidence not only from Fulton and DeKalb — where allegations were loudest — but also from Floyd, Fayette, and every other county with participating volunteers.

14.8 — What Actual Vote Would Have Contributed

With the Actual Vote app, volunteers video-record tabulator tapes — the precinct-level printouts generated by each scanner at the close of polls. The AV analysis then compares the totals on these tapes against the officially reported results. Any discrepancy between the two is flagged for investigation.

In Georgia 2020, this would have operated as follows:

Immediate detection of the memory card failures. In Floyd County, the scanner at the early voting location printed a tabulator tape showing the correct totals for all ballots it had processed — including the approximately 2,600 that were never re-scanned after the scanner jam. When the official results were released showing Floyd County’s vote totals, an AV comparison would have immediately identified that the official numbers were short of the tape totals by approximately 2,600 votes. The same logic applies to Fayette, Douglas, and Walton counties: the tapes showed the correct totals, but the official results did not match.

This comparison could have been performed within days of the election — well before the hand tally began on November 13. There would have been no need to wait for a statewide hand recount to discover that thousands of votes were missing. The cost would have been minimal: a volunteer with a smartphone at each affected precinct.

Independent preservation of Fulton County’s tape records. AV photographs would exist independently of whether poll workers signed the tapes, and independently of whether the tapes themselves were subsequently lost or destroyed. The 130 unsigned tapes and 10 missing tapes in Fulton County represent a gap in the official evidentiary record. AV photographs would have filled that gap — not by replacing the official chain of custody, but by providing a separate, timestamped record of what the tapes showed at the time they were generated.

Statewide evidence base. Perhaps most importantly, AV would have produced evidence from every county where volunteers participated — not just the counties that happened to attract political attention. This would have provided a basis for assessing reporting accuracy across Georgia, including in the 134 counties with no Carter Center monitors and the rural counties that received no particular post-election scrutiny.

Reduced ability to recycle allegations. Five years after the 2020 election, the same claims about Fulton County continue to generate investigations, litigation, and political controversy. An independent AV dataset — showing what every tabulator tape in the county actually said — would have provided a definitive reference point against which specific claims could be evaluated, rather than allowing allegations to persist in the absence of readily accessible evidence.

What AV Would Not Have Done

AV operates at the reporting layer — between tabulation at the precinct and the release of official results. It would not have resolved several other controversies that arose in Georgia 2020:

It would not have addressed concerns about whether ballot marking devices correctly recorded voter intent, particularly the ongoing debate about whether the QR codes on BMD-printed ballots could diverge from the plain-text summary that voters review.

It would not have verified voter eligibility, signature matching on absentee ballots, or the legitimacy of individual ballots.

It would not have detected errors that occurred during the counting process itself — for example, the double-scanning of ballots during the Fulton County machine recount, which was a counting-layer rather than reporting-layer issue.

And it would not have prevented the political pressure campaign on Georgia election officials, which was driven by claims that go far beyond anything that could be resolved by comparing tabulator tapes to official results.

Transparency and the deterrent effect in Georgia 2020

Georgia 2020 is the clearest illustration in this collection of what our methodology document calls the deterrent value of independent evidence. Three separate counting processes — the initial count, the statewide hand tally, and the machine recount — all confirmed the same outcome, and every formal investigation concluded that the documented procedural errors did not change the result. By the evidentiary standards of American elections, Georgia 2020 was an unusually well-verified election. And yet five years later the same debunked claims continue to resurface in new venues, because no body of independent precinct-level evidence exists that a skeptical citizen of any party can consult directly. Each new cycle of allegation forces the state and its administrators to re-litigate the same factual questions under political pressure, using evidence that is produced entirely by the officials whose conduct is being challenged. Had an Actual Vote corpus existed for Georgia 2020, every recycled claim would have had a publicly available, independently produced reference point against which it could be tested by anyone. This is the deterrent function of transparency: not that it prevents the claims from being made, but that it flattens the political cost of rebutting them and removes the rhetorical advantage that accrues to a claim no independent evidence can address. In an environment where verification authorities are distrusted by large fractions of the electorate regardless of their actual performance, evidence that does not route through any verification authority is the only kind that can do this work.

14.9 — Subsequent Events

The Georgia 2020 election generated an extraordinary volume of post-election litigation and investigation that continued for years afterward.

In May 2024, the State Election Board formally reprimanded Fulton County for procedural errors during the 2020 recount, citing more than 140 violations of election rules, including the duplication of ballot images and failures in data backup. The Board appointed an independent monitor for Fulton County’s 2024 elections. A consent order in June 2023 had previously resolved a related complaint, with state investigators concluding that data entry errors during the hand tally — including batches entered twice under different headings and misread numbers — were “not due to intentional misconduct by Fulton County elections staff” and “did not affect the result of the 2020 General Election.”

In December 2025, Fulton County’s admission regarding the unsigned tabulator tapes brought renewed attention to the 2020 election, as described above.

[FLAGGED FOR JASON’S REVIEW — the following section on the 2026 FBI raid may be cut. It is AV-relevant (illustrates how independent evidence prevents recycling of debunked claims) but is politically charged and involves ongoing litigation.]

On January 28, 2026, FBI agents raided the Fulton County election center and seized 2020 ballots and other materials. A subsequently unsealed affidavit showed the investigation originated from a referral by Kurt Olsen, a Trump-appointed official, and cited five “deficiencies or defects” as probable cause. At a federal court hearing in March 2026, election expert Ryan Macias testified that each of the cited defects was a mischaracterization of normal election processes: “missing” ballot images that were never part of the cast vote record, “duplicate” ballots that were standard remakes of damaged or digital originals, routine discrepancies between hand counts and machine counts during the RLA, unfolded absentee ballots that were military or overseas votes, and an internal interim tally mistaken for a final count.

The FBI and Department of Justice had previously investigated the same claims and concluded they were not supported by the evidence. Georgia’s Republican Secretary of State Raffensperger criticized the investigation. Nonpartisan election experts described the affidavit’s claims as recycled and debunked. As of April 2026, Fulton County has filed suit against the DOJ seeking the return of the seized materials, and the matter remains in litigation.

[END FLAGGED SECTION]

The longevity of these controversies underscores a point relevant to this case study. Five years after the election, the absence of readily accessible, independent evidence about what Georgia’s tabulator tapes actually showed has allowed the same allegations to cycle through investigations, lawsuits, political commentary, and now federal law enforcement action. Each new iteration requires fresh rounds of expert testimony and litigation to address claims that could have been evaluated against an independent evidence base from the beginning.

14.10 — What We Know and What We Don’t

What we know:

Joe Biden won Georgia by 11,779 votes, as confirmed by three separate counting processes. Nearly 6,000 votes across four counties were initially missing from the official results due to memory card upload failures — all reporting-layer errors discovered during the hand tally. More than 130 tabulator tapes in Fulton County were unsigned in violation of state rules, and 10 tapes are missing entirely. The hand tally itself, while it confirmed the correct winner, was conducted with significant procedural variation across counties, as documented by the Carter Center. Every formal investigation of the 2020 Fulton County results has concluded that errors were the product of human mistakes under time pressure, not intentional misconduct, and did not change the outcome.

What we don’t know:

Whether any additional reporting errors exist in the 134 counties that had no independent monitors during the hand tally. What, if anything, systematic evidence from every precinct would have revealed about reporting accuracy across the state. Whether the missing Fulton County tabulator tapes would have shown results consistent with the official totals. And whether routine, low-cost independent auditing of the reporting layer would have shortened the cycle of allegations, investigations, and litigation that has continued for more than five years.

14.11 — Why It Matters

Georgia 2020 is, by some measures, the most audited election in American history. Three full counts of the ballots, a signature audit, a Carter Center monitoring mission, multiple state investigations, a consent order, a formal reprimand — and, five years later, a federal raid and ongoing litigation.

And yet, with all of that scrutiny, the basic question at the heart of the Actual Vote system — do the numbers on the tabulator tapes match the officially reported results? — was never systematically answered for all 159 counties. The hand tally answered a different question (do the paper ballots match the machine count?) and answered it conclusively. But the reporting-layer question — whether the electronic transmission from precinct to county faithfully reflected what the scanners actually counted — was only resolved in the four counties where it happened to fail badly enough to be caught.

The memory card failures illustrate that reporting-layer errors are real, can affect thousands of votes, and can go undetected without either extraordinary measures (a full hand recount) or ordinary ones (a volunteer with a smartphone photographing a tabulator tape). The unsigned and missing tapes in Fulton County illustrate that the official chain-of-custody procedures are not always followed, and that when they fail, the evidentiary gap can persist for years.

Actual Vote addresses both of these vulnerabilities. A photograph of a tabulator tape is cheap to produce, easy to interpret, and independent of official procedures. It catches memory card failures the moment official results are released. It preserves the tape’s contents even if officials fail to sign or retain them. And it works everywhere — not just in the counties that attract political suspicion, but in every county where a volunteer shows up.

Georgia spent millions of dollars and years of institutional attention verifying the 2020 presidential election. An Actual Vote analysis, conducted by volunteers at each precinct on election night, would have provided comparable evidence about the reporting layer — faster, cheaper, and from every corner of the state.

14.12 — Further Reading

Cross-references in this collection

Case 12 — Prince William County, Virginia, 2020 — Same class of reporting-layer failure, but undetected for over a year because Virginia had no statewide hand tally.