Case 3 — Gaston County, 1998

Note on dating. The AC website’s “Why You Should Care” page dates this incident to 2000, as does the original case study file name in the AC corpus. However, the sole primary source — a University of North Carolina School of Government article titled “Would North Carolina Have Looked as Bad as Florida on Election Night 2000?” — states that the incident occurred in 1998: “In 1998, Gaston County was using direct-record electronic voting machines for the first time.” The article was published in the context of the 2000 Florida crisis and used the 1998 Gaston County incident as a prior example of what could go wrong. The website date should be corrected to 1998.

3.1 — Introduction

In 1998, Gaston County, North Carolina deployed direct-record electronic (DRE) voting machines for the first time. On election night, equipment failures caused approximately one-third of the county’s precincts to report zero votes, and a separate software bug caused the tabulators to stop counting at 32,000 votes. The chair of the county board of elections opened the recovery process to full public observation, computer experts retrieved the correct vote totals from the machines’ internal drives, and four close races were resolved using the recovered data. There were no election protests.

This case study occupies a unique position in the manual. Unlike the LBJ 1948 case (acknowledged fraud), the Florida 2000 case (unexplained reporting anomaly), or the Tammany Hall case (systematic corruption), the Gaston County case is a positive outcome story — a case where the system broke but was repaired through transparency and public observation. It is included not because it demonstrates a failure of election integrity, but because it demonstrates what success looks like when the principles behind Actual Vote — independent observation, public access to precinct-level evidence, transparent resolution of discrepancies — are applied in practice.

The evidence base for this case is thin. The sole primary source is a UNC School of Government article that provides a summary account without identifying the specific machines, the specific precincts, the specific races, or the officials involved. Web research did not surface any additional documentation. The case study is honest about this limitation.

3.2 — What Happened

The UNC School of Government article provides the following account. (All factual claims in this section are drawn from that source unless otherwise noted.)

Gaston County was using DRE voting machines for the first time in 1998. Before the election, officials discovered problems with the machines and applied an upgrade. The upgrade was necessary but cut into the time available for testing the machines for known bugs. The county proceeded to election day with less testing than would have been ideal.

On election night, the bugs manifested in two ways.

First, approximately one-third of Gaston County’s precincts reported zero votes. The cause was that the data cartridges — the removable media used to transfer vote totals from the DRE machines to the central tabulation system — were not functioning. The machines had recorded votes internally, but the mechanism for transmitting those totals to the central count had failed. This is a reporting-layer failure in the most literal sense: the votes were counted correctly by the machines, but the reported totals were zero because the data transfer mechanism broke.

Second, a separate bug in the tabulation system caused it to stop tallying votes when the cumulative total reached 32,000. This appears to be a software counter limitation — potentially a 15-bit integer boundary (2^15 = 32,768), though the UNC article does not specify the technical cause. Regardless of the specific mechanism, the effect was that the central tabulation system underreported the total number of votes cast in the county.

The result was that Gaston County’s election night returns were substantially wrong: roughly a third of precincts showed zero votes, and the countywide totals were capped at an artificially low number.

3.3 — The Recovery

The chair of the county board of elections made a decision that is the most important fact in this case study: the chair opened the recovery process to all who wished to observe.

Computer experts opened the DRE machines and retrieved the vote totals from the machines’ internal hard drives. The internal drives — which stored each voter’s selections independently of the data cartridges used for reporting — were intact and contained the correct data. The correct totals were retrieved, verified, and used to produce the official results.

Two statewide court of appeals races, one state House of Representatives race, and one local race were close enough that their outcomes were determined by the recovered data — meaning the results would have been wrong without the recovery effort.

Because the process was conducted transparently, with public observation, confidence in the recovered results was high. No election protests were filed.

3.4 — What Actual Vote Would Have Caught

The AV analysis of this case requires a technical caveat about DRE machines and poll tapes.

The poll tape question

DRE voting machines — unlike optical scan tabulators — do not scan physical ballots. Voters make their selections on a touchscreen (or similar interface), and the machine records the votes electronically. At the close of polls, DRE machines typically print a summary tape showing the total votes recorded for each candidate — functionally equivalent to the poll tape printed by an optical scan tabulator. This printed summary is the DRE equivalent of a poll tape.

The UNC article does not mention poll tapes. It says the correct data was on the machines’ “internal drives.” Whether the specific DRE machines used in 1998 Gaston County printed poll tapes at the close of polls is not documented in the available source. The AC website’s “Why You Should Care” page states that “the poll tapes would have printed with the correct vote totals, making it immediately clear that the reported totals were wrong.” This is a reasonable assumption — most DRE machines produce close-of-polls summary tapes — but it is an assumption, not a documented fact for this specific case.

If poll tapes were printed

If the Gaston County DRE machines did print summary tapes at the close of polls, the AV scenario is straightforward. The tapes would have shown the correct vote totals for each precinct. When the data cartridges failed and the central tabulation showed those precincts at zero votes, the discrepancy between the poll tape values (correct, nonzero) and the official report (zero) would have been immediately and publicly visible.

In this scenario, AV would not have changed the outcome — because the county resolved the problem correctly through its own transparent process. But AV would have provided the independent evidence faster. The county’s recovery required computer experts to physically open the machines and retrieve data from the internal drives — a process that took time and required specialized expertise. The poll tapes, had they been recorded by AV volunteers and uploaded to the public archive, would have provided immediate evidence that the reported zeroes were wrong, without requiring physical access to the machines’ internals.

If poll tapes were not printed

If the DRE machines did not print summary tapes — which is possible for some early DRE models — then AV’s specific methodology (video-recording poll tapes) would not have been applicable. In this scenario, the only record of the correct vote totals was on the machines’ internal drives, which required expert intervention to access. AV cannot video-record data that was not printed.

This scenario illustrates an important boundary condition for AV’s methodology: AV depends on the existence of a printed poll tape. In jurisdictions that use DRE machines without printed tapes (or in any situation where the tabulation equipment fails to print a tape), AV’s methodology cannot be applied. The manual’s recommendations (§10) include advocacy for universal poll tape printing and posting requirements, in part for precisely this reason.

The 32,000-vote cap

The tabulation bug that capped the countywide total at 32,000 is a different category of problem. This was not a precinct-level reporting failure but a central tabulation bug. AV’s precinct-level methodology would have caught it indirectly: if the sum of all AV-recorded precinct-level poll tape values exceeded 32,000 but the official countywide total was 32,000, the aggregate discrepancy would have been detectable through AV’s comparison analysis. The error would appear as a systematic shortfall in the official totals relative to the precinct-level evidence.

3.5 — Why It Matters

The Gaston County case matters for this manual for three reasons.

First, it is a positive outcome case. The equipment failed, the reporting was wrong, and the county fixed it — through transparency and public observation. The chair of the board of elections made the critical decision to open the process to the public. Computer experts retrieved the correct data. No protests were filed. This is what it looks like when the principles behind AV are applied: when the process is transparent and the evidence is accessible, problems can be identified and resolved with public confidence in the outcome. AV systematizes this approach — providing the independent evidence and the public archive that make transparent resolution possible at scale, rather than depending on individual decisions by individual officials.

Second, it demonstrates that equipment failures in the tabulation and reporting layers are not hypothetical. Gaston County’s DRE machines counted the votes correctly but failed to report them — the data cartridges did not function, and the tabulation software had a counting cap bug. These are real, documented equipment failures that produced substantially wrong election night returns. Four races were close enough to be determined by the recovered data. Without the recovery, those races would have been certified with incorrect results.

Third, the case illustrates the difference between a system that depends on official goodwill and one that provides structural assurance. Gaston County’s recovery worked because the board chair chose to open the process. In a county where the chair had different incentives — or where the chair simply didn’t think of it — the same equipment failure could have produced a different, less transparent resolution. AV does not depend on the goodwill of any official. It creates independent, publicly-held evidence that exists regardless of whether the local board of elections chooses to be transparent. The structural guarantee — evidence that cannot be suppressed by any party — is what distinguishes AV from a system that depends on individual officials making the right call.

3.6 — Why Gaston County’s Recovery Worked: The North Carolina Structural Advantage

The UNC School of Government article places the Gaston County incident in the broader context of why North Carolina did not experience a Florida-style meltdown in the 2000 election cycle. The article identifies several structural features of North Carolina’s election administration that made transparent recovery possible — features that Florida lacked.

North Carolina had a centralized State Board of Elections with authority to determine election outcomes and order new elections if necessary. County boards of elections were bipartisan bodies appointed by the state board (composed of three Democrats and two Republicans, or vice versa, depending on the governor’s party), not independently elected partisan officials. The state board’s protest and appeal procedures provided a single, clear process for resolving disputes — unlike Florida’s bifurcated system of simultaneous protests and contests in multiple courts.

This structural context matters for the Gaston County case. The board chair who opened the recovery to public observation was operating within a system designed for transparency: bipartisan appointment, state oversight, and clear procedural authority. The decision to invite public scrutiny was made easier by the fact that the chair had institutional support for doing so.

The UNC article’s conclusion identifies systemic vulnerabilities that could undermine these safeguards: funding cuts that overburdened county election staffs, shortened cycles for absentee ballots and ballot preparation, shortened cycles for equipment testing, and the challenge of finding experienced precinct workers. Gaston County’s data cartridge failure — which occurred in part because an upgrade cut into testing time — is a direct illustration of the equipment testing vulnerability.

AV’s relevance here is that it provides a safeguard that operates independently of these institutional structures. Even in jurisdictions without North Carolina’s centralized oversight — or in jurisdictions where the structural safeguards have been weakened by the vulnerabilities the UNC article identifies — AV creates independent, publicly-held evidence that can serve the same transparency function that Gaston County’s board chair provided through personal initiative.

3.7 — The Recurring Pattern: Data Transfer Failures After 1998

The Gaston County incident is not an isolated case. A 2010 Brennan Center for Justice report, “Voting System Failures: A Database Solution” by Lawrence Norden, documents strikingly similar failures occurring a full decade later in other jurisdictions — suggesting that the class of problem Gaston County experienced in 1998 was never systematically addressed.

Butler County, Ohio, March 2008. During the state primary, Ohio officials discovered that several votes were dropped from memory cards even though the final report stated the cards had been counted. A subsequent investigation determined that at least 1,000 votes were undercounted across nine of Ohio’s forty-four counties using Premier (formerly Diebold) touch screen or optical scan systems. The vendor, Premier, conceded that the root cause was a “logic error” in their server software that caused votes to be dropped when multiple memory cards were uploaded simultaneously. The New York Times noted in an editorial that Premier had subsequently notified more than thirty states using its systems about the problem. (Brennan Center, pp. 10–11)

Humboldt County, California, November 2008. A post-election “Transparency Project” — in which an independent scanner recounted every optical scan ballot — discovered that the voting system had failed to count approximately two hundred ballots. The first batch of absentee ballots scanned into the system, known as “deck zero,” had vanished from the totals before officials finished scanning all ballots and certified the results. Wired and Computerworld reported that the voting system vendor had been aware of the “deck zero” problem for years but had not notified the Election Assistance Commission, state election directors, or other relevant authorities. (Brennan Center, pp. 12–13)

The pattern across these three cases — Gaston County 1998, Butler County 2008, Humboldt County 2008 — is consistent: data transfer mechanisms between voting machines and central tabulation systems fail silently, producing zero or reduced vote counts for affected precincts, and the failures are not self-announcing without independent verification. In each case, the problem was discovered only because someone looked: Gaston County’s board chair invited public scrutiny, Butler County’s election director persistently investigated vendor claims, and Humboldt County’s clerk implemented a post-election transparency project using an independent scanner.

AV systematizes the “someone looking” step. Rather than depending on the initiative of individual officials — who may or may not choose to investigate, and who may or may not have the resources to do so — AV provides automatic, precinct-level verification that would flag any data transfer failure producing a discrepancy between the poll tape totals and the officially reported results. The fact that this class of failure persisted for at least a decade after Gaston County, across multiple vendors and multiple states, makes the case for systematic independent verification stronger, not weaker.

3.8 — Consistency Notes

Date discrepancy

The AC website’s “Why You Should Care” page (both current and v2) dates this incident to 2000. The UNC School of Government article, which is the sole primary source, dates it to 1998. The website should be corrected.

Poll tape assumption

The website states: “The poll tapes would have printed with the correct vote totals, making it immediately clear that the reported totals were wrong.” This is a reasonable inference but not documented in the source. The UNC article mentions internal drives, not poll tapes. Whether the specific DRE machines used in 1998 Gaston County produced printed summary tapes is unknown from the available evidence. The website language should be qualified to note that this is an inference based on typical DRE behavior, not a documented fact for this specific case. A more precise formulation might be: “If the machines produced poll tapes — as most DRE machines do — the tapes would have shown the correct vote totals, making the reporting failure immediately visible."

"Source code” vs. “vote data”

The UNC article says computer experts “retrieved the source code that showed the proper vote totals.” This is almost certainly a misuse of the term “source code” — source code is the programming instructions for the software, not the stored vote data. What the experts retrieved was the vote data from the internal drives. The AC case study and website descriptions correctly characterize this as retrieving “vote data” or “vote totals” rather than “source code.”

3.9 — Further Reading

Primary source

Gary O. Bartlett and Robert P. Joyce, “Would North Carolina Have Looked as Bad as Florida on Election Night 2000?” Popular Government, Fall 2002, published by the University of North Carolina School of Government. Describes the Gaston County incident (pp. 7–8) as part of a broader analysis of North Carolina election infrastructure in the context of the 2000 Florida crisis. The sole identified primary source for the Gaston County incident.

Corroborating source — recurring failure pattern

Lawrence Norden, “Voting System Failures: A Database Solution,” Brennan Center for Justice, 2010. Documents fourteen voting system failure case studies from 2002–2009. Establishes that the class of failure experienced in Gaston County recurred across multiple vendors and states for at least a decade.

Cross-references in this collection

Case 7 — Butler County, Ohio, 2008 — The same class of memory-card upload failure, this time triggered by a documented GEMS logic error and acknowledged by the vendor.
Case 9 — Shelby County, Tennessee, 2015 — The same GEMS upload bug, undiscovered seven years after the Butler County admission, surfaces again to drop 1,001 votes in a Tennessee county.
Case 4 — Florida 2000 — The companion case study about a state that did not have North Carolina’s structural advantages and produced a national crisis instead of a quiet local recovery.

Other (no specific URL identified)

America Counts, “Why You Should Care” (current version and v2 draft). Story #1 currently dated 2000 (should be 1998).
Verified Voting, Election Equipment Database. General information on DRE voting machines and poll tape printing capabilities.
No additional primary sources surfaced in web research for this specific incident. The incident appears to have been resolved locally without generating state-level investigation, news coverage, or legal proceedings that would have produced additional documentation.