Case 9 — Shelby County, 2015
9.1 — Introduction
Smith gave the photograph to a member of the county election commission. When county technicians investigated, they found the problem was not limited to one precinct. Votes were missing from multiple polling places — all in areas with large concentrations of Black voters. In total, 1,001 votes had been dropped from the election night count. A candidate who had lost a close race filed a lawsuit. The county’s Administrator of Elections retired under pressure. State officials called for a criminal investigation. And the public eventually learned that Shelby County had been running a version of vote tabulation software with a known vote-dropping bug — the same bug that had been discovered in Ohio seven years earlier, the subject of our Butler County case study.
This case occupies a unique position in our collection. In every other case study, the Actual Vote counterfactual is hypothetical: we describe what AV would have caught if it had been in use. Here, someone did what AV volunteers do — photographed a poll tape, compared it to official results, found a discrepancy — and the result was exactly what AV promises: independently preserved evidence that exposed a problem every official safeguard had missed. Bennie Smith, with one phone at one precinct, demonstrated the entire Actual Vote value proposition. The question this case raises is not whether the methodology works. It demonstrably does. The question is what would have happened if Smith had been able to cover not one precinct but all of them.
9.2 — Background: Shelby County and Its Elections
Shelby County, Tennessee, is home to Memphis, the largest city in the state and one of the largest majority-Black cities in the United States. The county is approximately 54% African-American and has been predominantly Democratic in its voting patterns for decades — typically 60-70% Democratic in countywide elections.
Since 2006, Shelby County had used Diebold AccuVote-TSX touchscreen voting machines for in-person voting. These were direct-recording electronic (DRE) machines — voters made their selections on a touchscreen, and the machine recorded the votes digitally. Critically, the AccuVote-TSX machines in Shelby County were paperless: they did not produce a voter-verified paper audit trail. As Bloomberg Businessweek noted, “most counties in the state don’t keep paper records of ballots, so there are no physical votes locked in a room somewhere, ready to be recounted.”
The machines were managed by the GEMS (Global Election Management System) central tabulator — a personal computer running Diebold software that sat in a windowless room at the county’s election headquarters. GEMS received data from the individual voting machines, aggregated it, and produced the official election results. At the close of polls, each precinct’s voting machines printed a summary tape showing the vote totals — the poll tape — which was posted publicly at the polling place. The memory cards from the machines were then sealed in security bags, driven to upload centers, and their data transmitted to GEMS over a closed network.
The transition from precinct-level totals to GEMS-reported results is the reporting layer — the step in the process where Actual Vote’s methodology operates. And it is exactly the step where Shelby County’s votes went missing.
Richard Holden served as the county’s Administrator of Elections, the official responsible for managing the election infrastructure and reporting results.
9.3 — The Pattern: Unexpected Results
The story begins not in 2015 but several years earlier. Following every major election since introducing computerized voting in 2006, Shelby County had been sued — often with plaintiffs alleging some version of election fraud through hacking or data manipulation. None of those lawsuits had been successful.
In the August 2010 county election, several well-known local candidates lost unexpectedly. Ten candidates filed suit in Shelby County Chancery Court, alleging “massive improprieties” and “tampering.” The lawsuit attracted attention but did not produce systemic change.
By 2014, the pattern had intensified. The county general election produced a complete Republican sweep of approximately ten countywide offices — in a county where registered Democrats significantly outnumber Republicans and where the African-American population constitutes a majority. Multiple losing candidates approached Bennie Smith, a Memphis-based financial analytics professional and software developer, asking him to examine whether their elections had been manipulated.
Smith was firmly skeptical. As he later recounted in multiple interviews, he believed the candidates had simply lost: they hadn’t knocked on enough doors, hadn’t targeted their outreach effectively, hadn’t run strong enough campaigns. His initial engagement was what he describes as a “consolation prize” — he told the candidates he didn’t believe them but would take a look at the election software as a courtesy.
What he found changed his mind entirely.
9.4 — The Discovery: Fraction Magic
Smith obtained documentation of the GEMS election management system through court cases (the documents had been made available through BlackBoxVoting.org’s legal efforts). Examining the GEMS specifications and database structure, he discovered something that he initially could not believe: the system contained a built-in feature called “weighted election” that allowed votes to be counted as fractions rather than whole numbers.
In a standard election, every vote counts as one. One voter, one ballot, one whole number added to a candidate’s total. The weighted election feature overrides this: it allows an administrator to assign fractional values to votes — 0.2, 0.4, 0.6, 1.5 — so that some votes count for more and some for less. The total number of voters still appears correct in the system’s records, but the distribution of votes among candidates can be altered to produce any desired outcome.
The feature had a legitimate stated purpose: certain specialized elections, such as property-rights votes where landowners’ votes are weighted by acreage, or corporate elections where board members have different voting power, use weighted voting by design. But the feature existed in the same software used for general public elections — elections where the principle of “one person, one vote” is foundational.
Smith, drawing on his programming background, built a proof-of-concept application he called “Fraction Magic” to test what the weighted election capability could do in practice. The results alarmed him. As he later described it in an interview: “I get to be God without His permission when I do that. I could just pick winners and losers with untold Godlike precision.”
The discovery was covered by Bloomberg Businessweek in 2016 and documented in a multi-part series on BlackBoxVoting.org. It remains one of the most significant findings in election security research: a vulnerability that is not a bug but a feature — designed, documented, and present in one of the most widely deployed election management systems in the United States.
But Smith also identified what he believed to be the one check on this vulnerability: the poll tapes. Whatever manipulation might occur at the central tabulation level, the poll tape is printed at the precinct before the data reaches the central tabulator. If the poll tape says one number and the central tabulator reports another, something happened between those two points — and the discrepancy demands explanation.
9.5 — One Photograph: October 8, 2015
On October 8, 2015, Memphis held its municipal general election. The marquee race was the mayoral contest between incumbent A.C. Wharton and city councilman Jim Strickland. Further down the ballot was the race for city clerk of court — a position overseeing the collection of approximately $5 million in traffic fines annually — where two-term city councilwoman Wanda Halbert, who is Black, was running against several other candidates.
Smith, who in his free time built voter-turnout models for local politicians, had bet a friend ten candy bars that the polling place at Unity Christian Church — Precinct 77-01, a predominantly African-American congregation located about a mile from Graceland — would have a big turnout. When the polls closed at 7 p.m., Smith was at Unity Christian. He photographed the poll tape posted in the church window with his BlackBerry. The tape, printed directly from the voting machines, showed 546 votes cast (Bloomberg’s figure; Smith has recalled 548 in later interviews — a minor discrepancy in the secondhand record).
About a week later, when Shelby County released its first precinct-by-precinct breakdown, Smith ran down the list to the one precinct where he knew the tally for sure. Unity Christian showed 330 votes.
9.6 — What the County Found — And Didn’t Disclose
Smith showed the poll tape photograph to Norma Lester, a friend who served on the county election commission, on October 17 — the week before commissioners were scheduled to certify the election. The photograph was, as Bloomberg Businessweek described it, “smoking-gun proof that the votes had disappeared somewhere between Unity Christian’s voting machines and the GEMS tabulator that spat out the official election night tally.”
What happened next is documented through internal emails and records that were later obtained through open records requests by local attorney Carol Chumney and provided to Bloomberg Businessweek.
Joe Young, Shelby County’s deputy administrator of elections, went looking for answers. He examined server logs and other data showing how GEMS had operated on election night. The problem was much worse than one precinct. Young initially found votes missing from approximately 20 precincts. After further investigation, he confirmed missing uploads from voting machines at four polling places — all in areas with large concentrations of Black voters. Three were at Black churches: Mt. Zion Missionary Baptist, Mississippi Boulevard Christian, and Unity Christian. The fourth was at Gaston Park, a community center in a predominantly Black neighborhood south of downtown.
In total, according to a master tabulation record, 1,001 votes had been dropped from the election night count, including almost 400 from an early voting center at Mt. Zion — the most from any single polling place. Seven voting machines across these locations had failed to upload their data to GEMS. The system showed a specific and insidious behavior: when officials tried to reload the missing data from the memory cards on October 19, GEMS recognized at least some of the votes as already counted — even though it was clear from Smith’s poll tape and Young’s own investigation that they had not been.
Young was not rewarded for his diligence. On October 19, he received a sharp rebuke from his boss, Administrator Richard Holden, who had asked Young to investigate a separate issue. Discussion of the missing votes stops in the internal emails after that. There is no indication the county looked further into how the votes were lost, whether it had occurred in past elections, or what other elections over the preceding years might have been compromised.
Critically, the five county election commissioners were not told the full scope of the problem. As far as they knew, the only precinct with missing votes was Unity Christian. Commissioner Anthony Tate, one of two Democrats on the panel, later told Bloomberg Businessweek that if he had been informed of the full scale of the problem — four polling places, seven machines, 1,001 missing votes — he would not have voted to certify the election. The commissioners certified the results on October 23.
The county did add the missing votes after the fact, using data on the memory cards that Young located. Approximately 300 votes were added on October 19 and an additional 700 or so over the next two weeks. In the clerk of court race, the missing votes were divided among four African-American candidates, including Halbert, who gained 225 votes on the winner — not enough to change the result. She had lost by 1,367 votes.
9.7 — The Lawsuit
Wanda Halbert filed a lawsuit alleging that problems with Shelby County’s election systems “date back at least a decade and are pervasive, severe, chronic, and persistent.” In early filings, the county conceded that some of the software it uses is so old or obscure it doesn’t even know who makes it.
The case initially went before District Court Judge Jim Kyle. In May 2016, Judge Kyle abruptly recused himself, citing that his wife was running for office in November. A new judge was appointed in July 2016. Halbert, who had left her City Council seat to run for clerk, was by then cleaning houses to fund her lawsuit. Her attorney was a personal injury lawyer — the only one she could afford. The county was represented by John Ryder, a local attorney who also served as general counsel for the Republican National Committee, one of the most powerful election lawyers in the country.
The outcome of the lawsuit is not well documented in available public sources, but subsequent lawsuits filed by losing candidates in Shelby County similarly failed to produce systemic change.
9.8 — The Probable Cause: A Familiar Bug
Among the documents obtained through Chumney’s open records requests was a user’s manual for Shelby County’s version of GEMS. It revealed something that connects this case directly to our Butler County case study: Shelby County was running a version of GEMS that contained the vote-dropping bug discovered in Ohio in 2008.
The Butler County case documented how a logic error in GEMS caused votes to be silently lost during the upload of memory cards to the central tabulation server. When multiple cards were processed simultaneously, a sharing violation could occur, causing incoming vote data to overwrite other data still being processed. The votes on the displaced card were simply lost. GEMS showed the uploads as successful. No error messages appeared. The official results were wrong, and nothing in the system indicated a problem.
After a five-month confrontation between Butler County officials and the vendor (then called Premier Election Solutions, formerly Diebold), controlled simulations proved the bug was in the GEMS code itself — not, as the vendor had initially claimed, caused by antivirus software. Premier acknowledged the bug had been present for approximately ten years and notified roughly 1,750 jurisdictions across 34 states. A fixed version was developed.
But Diebold did not replace the flawed software versions outside of Ohio. For counties to upgrade on their own was expensive. Some counties in Virginia, Georgia, and Tennessee continued using the version with the known bug. Those in Virginia and Georgia employed special protocols to catch vote-dropping — including comparing tabulated results with physical poll tapes at every precinct.
Tennessee law required counties to do the same. But Shelby County had stopped the practice several years before the 2015 election to save money, according to a deposition in an earlier election lawsuit. The comparison was replaced with an audit of 10 percent of polling places — a sampling that failed to catch the problem in October 2015. The county’s election commissioners were also not told that Shelby County was using a version of GEMS known to drop votes, according to Commissioner Tate.
Whether the explanation is nefarious or merely negligent, the result was the same: 1,001 votes — disproportionately from Black precincts — disappeared from the official count, and no one in the election administration flagged the problem. It took a citizen with a camera phone to expose it.
9.9 — Richard Holden Retires
For years before the 2015 controversy, Administrator of Elections Richard Holden had faced criticism and pressure to resign. Multiple parties had expressed no confidence in his leadership. Calls for his removal had been persistent. He had resisted all of them.
On December 7, 2015 — approximately two months after Smith’s poll tape discovery — Holden appeared at a special election commission meeting called to certify results from a November 19 city council runoff. He announced his retirement, effective at the end of December. According to local media coverage, state officials called for a criminal investigation into Holden’s activities within minutes of his announcement. State Representative G.A. Hardaway was among those pressing for accountability.
No criminal charges against Holden are documented in public sources. A Memphis Flyer article headlined “Election Administrator Holden Off the Hook, but Critics Remain Unsatisfied” exists, but it concerns an earlier, separate episode — Holden’s completion of a probation period imposed after the 2012 wrong-ballot scandal, not the criminal investigation called for in December 2015. Whether that later investigation reached any formal conclusion has not been reported in available public records. Linda Phillips was appointed as Holden’s replacement. In a local television interview, Phillips said the current system needed only small changes, mostly voter education about confusing ballots.
The temporal sequence is clear: years of election anomalies, years of pressure to resign, consistent resistance — and then retirement within weeks of evidence surfacing that 1,001 votes had been dropped from the most recent election. We note the sequence without asserting that it proves deliberate wrongdoing. What it does establish is that the poll tape evidence created a degree of accountability that years of complaints, lawsuits, and institutional pressure had failed to produce.
9.10 — What We Don’t Know
Intellectual honesty requires acknowledging several gaps in what the public record tells us about this case.
The most likely explanation for the 2015 vote loss is the known GEMS bug — the same software flaw documented in our Butler County case study. The county was running the affected version; the symptoms match; an expert in GEMS software called it the probable cause. But “most likely” is not “confirmed.” Shelby County did not conduct a public investigation into the root cause. The internal investigation by Joe Young was cut short after Holden’s rebuke. The county cited Halbert’s pending lawsuit as the reason it would not comment on anything that happened election night.
We do not know whether the GEMS bug was the sole cause, or whether other factors — including the possibility of deliberate manipulation — contributed. Smith’s separate discovery of the Fraction Magic weighted-election vulnerability in GEMS raises the question of whether that feature was ever activated in Shelby County. The circumstantial connection is notable: Smith found both the Fraction Magic capability and the vote discrepancy in the same county using the same system. But the Bloomberg Businessweek investigation, which had access to internal county documents, pointed to the upload bug rather than Fraction Magic as the probable mechanism. These are two separate vulnerabilities in the same software — one a bug, one a feature — and the available evidence does not conclusively establish which was at work, or whether both played a role.
We do not know the full scope of the problem. Smith references 22 or more precincts with anomalies in his public accounts. Young’s internal investigation initially found approximately 20 precincts affected, then narrowed to four polling places with confirmed missing uploads. The discrepancy may reflect different criteria: Smith may be counting precincts with any statistical anomaly, while Young was counting confirmed missing machine uploads. In either case, the problem extended well beyond Unity Christian.
We do not know whether previous elections were similarly affected. Shelby County had been using the same GEMS version for years. The vote-dropping bug is triggered by specific conditions during upload — conditions that could occur in any election. If the county was not comparing poll tapes to official results (having stopped the practice to save money), there would have been no mechanism to detect vote loss in prior cycles. The question of how many past elections were compromised by the same bug is unanswerable from available evidence, and no one in authority appears to have tried to answer it.
9.11 — The Broader Vulnerability: Fraction Magic
While the 2015 vote loss was most likely caused by the known upload bug, Smith’s separate discovery of the weighted election feature in GEMS deserves attention in its own right. This was not a bug but a designed feature — documented in the vendor’s own specifications — that allowed votes to be counted as fractions.
Bloomberg Businessweek covered the discovery in its 2016 article, and Smith documented it in detail through the Fraction Magic proof-of-concept and a multi-part series on BlackBoxVoting.org. The weighted election capability was not unique to Shelby County. GEMS was one of the most widely deployed election management systems in the United States, in use across hundreds of jurisdictions in dozens of states.
The vendor’s stated justification — that the feature supported legitimate weighted voting in property-rights and corporate elections — raises a structural question: why would a feature with no legitimate purpose in public “one person, one vote” elections be present in the software used for those elections? As Smith observed in interviews, election software should count votes as whole numbers and nothing else. The capacity for fractional counting is either unnecessary or dangerous, and in either case, its presence in public election systems demands explanation.
Whether the weighted election feature was ever activated in Shelby County or any other jurisdiction to manipulate a public election is not established. What is established is that the capability existed — documented, designed, and ready to use — in systems that counted millions of votes in American elections. (See the dedicated Fraction Magic Attack case study for full technical detail.)
9.12 — 2018: The Outcome Changes
In the elections following Smith’s discovery and Holden’s departure, Shelby County’s results changed dramatically.
Most strikingly, Wanda Halbert — the candidate whose 2015 race was directly affected by the missing votes, who had filed the lawsuit, who had been cleaning houses to fund her legal fight — won election as Shelby County Clerk in 2018. Floyd Bonner was elected as the first African-American sheriff of Shelby County. Multiple Black candidates won countywide offices. The composition of elected county leadership shifted to more closely resemble the demographics of the electorate.
Smith and others have drawn a direct connection: when the election reporting system came under scrutiny, the results began to reflect the actual will of the voters. The same voters, in the same precincts, with the same demographic composition, produced different outcomes once the reporting process was subject to independent oversight.
This inference is reasonable but must be stated carefully. Other factors could contribute to the 2018 shift: the national political environment (a strong Democratic wave year), changes in candidate quality and campaign strategy, increased voter mobilization, and the general heightening of attention to election integrity. We cannot isolate the effect of the 2015 discovery from these other variables.
What we can say is this: for years, Shelby County’s election results defied the demographic and political expectations that its population would predict. After a citizen produced independent evidence of a massive reporting discrepancy — and after the administrator responsible for reporting retired and was replaced — the results shifted to match those expectations. The most parsimonious explanation is that something about the reporting process changed. The most visible change in the reporting process was that it was now under scrutiny.
9.13 — What Actual Vote Would Have Contributed
What AV would have caught
Exactly what Bennie Smith caught — and more. The discrepancy between poll tape totals and officially reported results is precisely the reporting-layer problem that Actual Vote is designed to detect. If AV volunteers had been present at Shelby County precincts on election night in October 2015, they would have photographed poll tapes and uploaded the data to AV’s independent server. When the official results were released, the comparison would have revealed every precinct where the official numbers didn’t match the poll tapes.
Smith covered one precinct. AV would have covered all of them. The 1,001 missing votes across multiple polling places would have been flagged simultaneously — not discovered weeks later through a deputy administrator’s quiet investigation that was shut down by his boss, and not dependent on a single citizen’s analytical skills and initiative.
What AV provides beyond what Smith accomplished
The difference between Smith’s individual effort and what AV offers is the difference between a proof of concept and a deployed system.
Scale matters. Smith photographed one poll tape and found a 40% discrepancy at one precinct. The full scope — four polling places, seven machines, 1,001 missing votes — was discovered only because his photograph forced an investigation. But that investigation was internal, conducted by county staff, and its findings were not fully shared with the commissioners who had to certify the election. With AV, the discrepancies at every affected precinct would have been independently documented and publicly visible on election night — before the officials under scrutiny had any opportunity to decide what to investigate, what to disclose, and what to leave alone.
Speed matters. Smith discovered the discrepancy about a week after the election, when precinct-level results were released. AV’s process of photographing poll tapes and uploading them to an independent server preserves the precinct-level data within hours of polls closing. Discrepancies can be identified as soon as official results are published — not days later, and not dependent on when a citizen happens to check the numbers.
Independence matters. The internal emails show that when Joe Young investigated and found the full scope of the problem, his boss rebuked him and the discussion stopped. The county’s five commissioners were told about one precinct, not four. The officials responsible for the reporting system controlled the flow of information about what went wrong with the reporting system. AV removes this dependency. Poll tape photographs uploaded to AV’s server exist outside the election administration’s control. They cannot be suppressed, minimized, or selectively disclosed.
Redundancy matters. Smith was one person. If he had chosen a different precinct, or if Unity Christian had happened to be one of the precincts where the upload succeeded, the discrepancy might never have been found. AV’s volunteer model means multiple people photograph multiple tapes across every precinct, and the system doesn’t depend on any single individual’s presence or luck.
What AV would not have caught
AV detects reporting-layer discrepancies — differences between the precinct-level count and the officially reported results. It does not detect counting-layer problems: errors or manipulation that occur before the poll tape is printed.
The AccuVote-TSX machines in Shelby County had documented counting-layer problems. During the 2016 federal election, touchscreen calibration issues caused votes for one presidential candidate to register as votes for another on the machine itself. Because the machines are paperless DREs, there is no independent record of voter intent to compare against the machine’s tally. If a machine miscounts votes, the poll tape reflects the incorrect count, and AV would show no discrepancy between the wrong poll tape and the equally wrong official results.
AV also cannot determine the cause of a discrepancy it detects. If AV had been deployed in October 2015, it would have flagged the vote discrepancies immediately. But it would not have told anyone whether the cause was the GEMS upload bug, the Fraction Magic vulnerability, or something else. AV identifies that something went wrong in the reporting layer. Determining why requires investigation — ideally, the kind of thorough, public investigation that Shelby County did not conduct.
The proof-of-concept point
This bears repeating because it is the central lesson of this case: Bennie Smith’s experience is not a hypothetical Actual Vote scenario. It is the Actual Vote methodology, executed by one person at one precinct with one phone. The result was evidence sufficient for a court to act — court-quality data, independently sourced, preserving a record that the official system failed to preserve.
Every other case study in this collection asks: what would have happened if AV had been there? This case answers a different question: what happens when someone does what AV does? The answer is: they catch a 40% vote discrepancy that every official safeguard missed, they produce evidence that enables judicial action, and the accountability that follows changes the trajectory of elections in their community.
AV’s contribution is not to invent this methodology. It is to make it available to everyone, everywhere, systematically — so that the next community doesn’t need to produce its own Bennie Smith in order to know whether its votes were counted accurately.
Seven years after Smith’s discovery, in Monmouth County, New Jersey, a citizen named James Bean noticed that 311 more votes had been counted than cast in his local mayoral race. He filed a records request. He was told he was comparing “apples to oranges.” It took over two months before anyone took him seriously. Bean’s experience — persistence rewarded, but slowly and painfully — shows what happens when AV’s methodology is applied without AV’s infrastructure. Smith had a photograph and technical expertise. Bean had a records request and weeks of unanswered emails. Both found real errors. Neither should have had to work that hard.
9.14 — The Cross-Reference: Butler County Revisited
This case has a direct connection to our Butler County, Ohio case study that deserves explicit attention.
In 2008, Butler County officials discovered that GEMS was silently dropping votes during memory card uploads. After a five-month confrontation with the vendor, controlled simulations proved the bug was in the GEMS code. The vendor acknowledged it had been present for approximately a decade and notified 1,750 jurisdictions across 34 states. A fix was developed.
Seven years later, in 2015, the same bug — or a bug producing identical symptoms — dropped 1,001 votes in Shelby County. The county was running a version of GEMS known to contain the flaw. The vendor had not replaced the software outside Ohio. Shelby County had stopped performing the poll-tape-to-official-results comparison that Tennessee law required and that would have caught the problem.
The Butler County case study asks: what happens when a software bug silently drops votes for a decade across 34 states? The Shelby County case study provides part of the answer: it keeps dropping them. The fix that was developed in Ohio in 2008 didn’t reach Memphis by 2015. Betty McGary’s persistence in Butler County saved future Ohio elections from the bug. It did not save Shelby County.
This is the systemic problem that Actual Vote addresses. Fixes that depend on vendors, state mandates, or official diligence can fail to propagate. A bug that is known, documented, and fixed in one jurisdiction can continue causing harm in another for years. Independent verification by citizens — the kind Smith performed, the kind AV systematizes — operates outside these institutional dependencies. It catches the problem wherever it occurs, regardless of whether the jurisdiction has updated its software, maintained its audit protocols, or been told by its vendor that the bug exists.
9.15 — Why It Matters
The Shelby County case demonstrates four things that matter for anyone considering whether independent vote reporting verification is worth the effort.
First, reporting-layer failures of enormous magnitude can go entirely undetected by official processes. One thousand and one votes disappeared from the election night count, and no one in the election administration flagged the problem — not the GEMS system, not the election night technician, not the administrator, not the canvassing process. The discrepancy was caught by a private citizen with a camera phone. When a deputy administrator investigated and found the full scope, he was rebuked by his boss, and the findings were not shared with the commissioners responsible for certifying the election. Without Smith, those 1,001 votes would have simply been missing from the permanent record — and no one would have known.
Second, independent evidence works. Smith’s photograph — a picture of a publicly posted document, taken with a BlackBerry — was sufficient to force an investigation and, ultimately, a lawsuit. It was not ambiguous, not subject to interpretation, not dependent on institutional cooperation. The poll tape said 546. The official results said 330. The gap required explanation. This is exactly the kind of evidence AV produces: independently sourced, timestamped, precinct-specific, and conclusive about the existence of a discrepancy even when the cause requires further investigation.
Third, known vulnerabilities persist when fixes depend on institutional diligence. The GEMS bug was discovered in 2008, acknowledged by the vendor, and fixed in Ohio. Seven years later, it was still dropping votes in Tennessee — because the vendor didn’t push the fix, the county didn’t demand it, and the state didn’t enforce the poll-tape comparison requirement that would have caught the problem. AV operates independently of all these institutional dependencies.
Fourth, the scale problem is real. Smith’s individual effort exposed the problem at one precinct. The full scope — four polling places, seven machines, 1,001 votes — emerged only through a county investigation that was shut down by the administrator and whose findings were partially concealed from the certifying commissioners. How many other elections — 2006, 2008, 2010, 2012, 2014 — experienced the same bug dropping the same votes with no one checking? The question is unanswerable, and that unanswerable quality is itself the problem AV exists to solve.
The arc of this story — from Smith’s skepticism to his discovery, from Halbert’s loss to her eventual election as Shelby County Clerk in 2018, from Holden’s resistance to his retirement — also illustrates what transparency produces. Not conspiracy theories. Not partisan warfare. Evidence-driven accountability that ultimately led to elections more reflective of the community that cast the ballots. As Reverend Eric Lowell Winston, pastor at Mt. Zion Missionary Baptist Church (where the most votes were lost), told Bloomberg Businessweek about the absence of clear answers: “I think it insults the intelligence of our community. As if we don’t see or understand. But we do understand.”
Actual Vote exists so that every community has the evidence to understand — and the evidence to act.
9.16 — Further Reading
Primary investigative source
- Michael Riley, Jordan Robertson, and David Kocieniewski, “The Computer Voting Revolution Is Already Crappy, Buggy, and Obsolete,” Bloomberg Businessweek, September 29, 2016. The definitive contemporaneous reporting on the Shelby County 2015 incident, drawing on internal Shelby County election commission emails and documents obtained by Carol Chumney through open records requests.
Bennie Smith — additional materials
- “Scrutineers Series, Part II — Bennie Smith,” WhoWhatWhy podcast, 2020.
- “About Bennie Smith,” benniesmith.com.
Fraction Magic series
- Bev Harris and Bennie Smith, “Fraction Magic – Part 1: Votes are being counted as fractions instead of as whole numbers,” BlackBoxVoting.org, May 12, 2016 (Parts 1–6 with supporting source code analysis).
Cross-references in this collection
- Case 7 — Butler County, Ohio, 2008 — The original discovery and vendor admission of the GEMS upload bug that almost certainly caused the Shelby County 2015 vote loss.
- Case 11 — Fraction Magic Attack — Full technical treatment of the architectural feature in GEMS that Smith uncovered while investigating Shelby County.
- Case 16 — Monmouth County, New Jersey, 2022 — Citizen detection of a reporting-layer error through records request rather than poll-tape photograph; illustrates what AV’s methodology looks like without AV’s infrastructure.
Original Sources list (preserved)
- Action News 5, “Shelby County election commissioner announces retirement” (December 7, 2015)
- Memphis Daily News, “Holden Calls It Quits As Shelby County Election Administrator” (December 7, 2015)
- Memphis Flyer, “Election Administrator Holden Off the Hook, but Critics Remain Unsatisfied”
- Truthout, “Ten Candidates File Suit as ‘Massive Improprieties, Tampering’ Seen in Shelby County, Tennessee, Election” (re: August 2010 election)
- Bennie J. Smith, “Hand-marked Paper Ballots: How this Tried-and-True Method Makes Us More Secure,” The Commercial Appeal (November 21, 2019)
- Shelby County Election Commission, history of election equipment (electionsshelbytn.gov)
- Local media coverage of 2018 Shelby County election results, including Halbert’s election as Shelby County Clerk (Ballotpedia, Action News 5, Fox 13 Memphis, WREG)