Case 17 — DeKalb County, 2022

17.1 — Introduction

On the evening of May 24, 2022, Michelle Long Spears drove to the precinct where she and her husband had voted earlier that day. Spears was a candidate in the Democratic primary for the District 2 seat on the DeKalb County Board of Commissioners. She and her husband had each cast a ballot for her. The poll tape posted outside the precinct that night showed her with zero votes.

She drove to other precincts in the district. The poll tapes there showed the same thing: zero votes for Spears. By the morning of May 25, her campaign team had photographed more than a dozen poll tapes, all reporting Election Day vote totals for Spears that ranged from zero to a small handful — even in precincts where her support, by every other indicator, had been substantial. The unofficial overnight summary showed her in third place with 24% of the vote, behind Marshall Orson and Lauren Alexander.

After a hand count of all 40 District 2 precincts conducted over Memorial Day weekend, the picture changed completely. Spears had received not zero Election Day votes but 4,078. She finished first overall with 43% of the vote, qualified for the June 21 runoff, and went on to win it with 61%. Marshall Orson, the apparent second-place finisher on election night, dropped to third in the certified hand count and was eliminated from the runoff.

The cause was a programming error that had nothing to do with malice or fraud. After the original project was built, a candidate withdrew, requiring a change. Then five precincts were discovered to have been redistricted into District 2 mid-cycle, requiring a separate parallel election project. A text-display problem on the ballot marking devices required yet another change. When the Secretary of State’s Center for Election Systems pushed the third change, it was built from the wrong base copy, and the result was a mismatch between what the touchscreen ballot marking devices showed (three candidates) and what the precinct scanners expected (four candidates). At 32 of 40 District 2 precincts, the scanners read every ballot one position off.

This case study occupies an unusual position in the collection. The error described here is, strictly speaking, a counting-layer failure rather than a reporting-layer failure: the precinct scanners produced poll tapes that already contained the wrong numbers. In most counting-layer cases, Actual Vote’s standard workflow — comparing publicly posted poll tapes against officially reported results — would not detect a discrepancy, because both numbers reflect the same incorrect count. This case is the exception. The error here was so severe and so specific — a candidate showing zero votes in precincts where she demonstrably had them, including her own precinct — that the poll tape itself was identifiable as wrong by anyone with knowledge of the ground truth, beginning with the candidate. AV’s poll-tape transparency caught it not because the tape and the reported result diverged, but because the tape was visibly impossible.

This is also the textbook “For Candidates” case in the collection. Spears performed, by hand and under enormous time pressure, exactly the workflow Actual Vote is designed to systematize: visit precincts, photograph poll tapes, aggregate results, identify anomalies. She did it because no other tool existed. The case is a direct argument for why one should.

17.2 — Background: How DeKalb County Reports Votes

DeKalb County is the fourth-largest county in Georgia, with approximately 760,000 residents and roughly 540,000 registered voters. It encompasses the eastern portion of metropolitan Atlanta, including the cities of Decatur, Brookhaven, Stone Mountain, and Tucker. The county is heavily Democratic in its voting patterns and majority-minority in composition.

Like every Georgia county, DeKalb conducts its elections on equipment manufactured by Dominion Voting Systems and standardized statewide following Georgia’s 2019 voting-system overhaul. Voters cast ballots using ImageCast X (ICX) ballot marking devices — touchscreen units that record voter selections digitally and then print a paper ballot showing the selections. The paper ballot is fed into an ImageCast Precinct (ICP) scanner, which reads the printed selections and tabulates the vote. The two pieces of equipment are conceptually distinct: the touchscreen records voter intent and prints a paper artifact; the scanner reads the printed artifact and produces a numerical tally.

Both pieces of equipment are programmed with election-specific configurations. Before each election, the Georgia Secretary of State’s Center for Election Systems (CES) — a state-level office that handles ballot programming for Georgia counties — builds an “election project” for the county. The project specifies what races appear on which ballots, in what order candidates are listed, what positional slots each candidate occupies on the touchscreen and on the scanner’s reading map, and what district boundaries determine which voters see which races. The project is loaded onto the county’s election management server, then distributed to BMDs and scanners through memory cards.

Georgia is one of the more transparent states for poll-tape access. State law and Secretary of State guidance require that the precinct-level summary tape be printed at the close of polls and posted publicly at the polling place — meaning any voter, candidate, journalist, or election observer can photograph the tape on election night without requesting access from any official. This is the access model that makes Actual Vote’s workflow most natural to deploy: the primary evidence is publicly available in real time.

DeKalb’s District 2 contains approximately 40 precincts, distributed across the inner-eastern parts of the county including portions of Decatur, Druid Hills, Avondale Estates, and surrounding areas.

17.3 — The Cascade

The technical cause of the May 24 error is documented in detail in DeKalb VRE’s May 24 Commissioner D2 Analysis, released July 13, 2022. It is a four-step cascade, each step individually reasonable, the cumulative effect catastrophic.

Step 1 — The original project. When CES built the original election project for the May 24 primary, the District 2 race had four candidates: Lauren Alexander, Donald Broussard, Marshall Orson, and Michelle Long Spears, listed alphabetically. The BMDs displayed all four candidates. The scanners were configured to read four candidate positions per ballot in the District 2 race.

Step 2 — The candidate withdrawal. After the election project had been built and equipment had been prepared, Donald Broussard withdrew from the District 2 race. DeKalb VRE marked Broussard’s status as “disabled” in the county’s copy of the election management server. This is the proper procedure: it preserves the historical record of the project while preventing votes from being credited to a candidate who has withdrawn. Any votes cast for Broussard after this change — for example, on absentee ballots that had already been mailed — would be counted as blank rather than as votes for Broussard.

Step 3 — The redistricting fix. Subsequent to the withdrawal, it was discovered that Georgia’s post-2020 redistricting, delayed by the pandemic, had been finalized late. DeKalb was the last Georgia county to settle its new district boundaries. DeKalb VRE discovered that 6,716 voters across five precincts had been assigned to the wrong County Commission district. To correct this, CES built a new, parallel election project for those five precincts so that District 2 properly appeared on their ballots. To accommodate voters from those five precincts during early voting, each of DeKalb’s 14 advance voting locations was equipped with at least one designated BMD and one designated scanner running the parallel project; the Emory campus advance site had two designated BMDs.

Step 4 — The text fix. On the first day of early voting, an unrelated display issue was discovered: the text of one Republican party question was not appearing properly on the BMDs. CES updated the election project to correct the text. The update was built using the county-maintained copy of the election project — the copy in which Broussard had been disabled. This was a procedural mistake. When the text-fix update was deployed to BMDs across DeKalb on Election Day, it overwrote the touchscreen ballot configuration in 32 of the 40 District 2 precincts. The BMDs in those precincts now displayed three candidates in the District 2 race (Alexander, Orson, Spears) — Broussard’s slot having been removed by the disabled-status flag in the county copy. But the scanners in those same 32 precincts had not been updated. They still expected four candidate positions, with Broussard occupying the second slot.

The result was a positional shift. When a voter selected Spears (third position on the three-candidate BMD), the scanner read the selection in its third position — which on the four-candidate map was Marshall Orson. When a voter selected Orson (second position on the three-candidate BMD), the scanner read it in its second position — which was Broussard’s disabled slot, and so the vote was recorded as a blank. Lauren Alexander, who appeared first on both configurations, was unaffected. The DeKalb analysis document calls this the “ballot shift”: every vote below Alexander was shifted upward by one position in the scanner’s reading.

The PDF documents four distinct configurations across the 40 District 2 precincts on Election Day:

Four precincts had matching BMD-and-scanner configurations and no errors. These were the correctly-configured precincts.
Four precincts had BMDs that still displayed Broussard’s name (the BMD update had not been applied) and scanners that still expected four candidates. Voters who selected Broussard had their votes recorded as blank when memory cards were uploaded, because of his disabled status, but no positional shift occurred. Spears, Orson, and Alexander were all counted correctly.
Twenty-nine precincts had three-candidate BMDs and four-candidate scanners — the shifted configuration. Spears votes were recorded as Orson; Orson votes were recorded as Broussard (and then converted to blank on upload).
Three precincts had mixed configurations. Scott and Shamrock each operated two scanners, one correctly configured and one shifted. Lakeside experienced a separate technical failure that required some ballots to be brought back to DeKalb VRE’s offices for rescanning; the ballots scanned at the polling place were shifted, while the ballots scanned at the office were not.

The shift affected only Election Day votes cast on equipment running the affected configurations. Advance voting and absentee ballots — which had been processed on the designated BMDs and scanners running the parallel project, or on absentee scanners at the central office — were unaffected.

17.4 — Election Night and the Zeros

DeKalb posts poll tapes publicly outside polling places at the close of voting, as Georgia’s transparency framework requires. On the evening of May 24, observers and candidate representatives at multiple District 2 precincts began noticing something odd: the tapes showed Michelle Long Spears with zero Election Day votes — not low totals, but exact zeros — at precinct after precinct.

Spears’s campaign team reached her with the news during the evening. Her own precinct, where she and her husband had voted that morning, was among the affected ones. By her own account, given in public testimony at a May 31 special meeting of the DeKalb Board of Registration and Elections: “I drove to my own precinct where my husband and I voted for myself and there were zero votes for me. In the end, I collected over a dozen photos of precincts with zero votes for me on Election Day. I sounded the alarm and we let the DeKalb VRE know about these discrepancies.”

The campaign brought the photos to media on Monday, May 30 — day two of the hand count. Local commissioners Jeff Rader and Ted Terry began publicly raising questions.

The unofficial overnight numbers told the story algorithmically. The Election Summary Report from “Database 1” (covering the 39 precincts in the main report, posted at 2:30 a.m. on May 25) showed Spears with 3,031 total votes and 23.98% of the vote — running third behind Orson and Alexander. A separate “Database 2” report (covering the four redistricted precincts) showed Spears with 639 votes (44.87%) — running first by a wide margin in those precincts. The contrast between the two databases was itself diagnostic: in the precincts running the parallel project (Database 2), Spears was demonstrably the leading candidate; in the precincts running the affected project (Database 1), she was apparently in third. No legitimate dynamic of voter behavior could produce that pattern. Some technical artifact had to be involved.

But on the public-facing summary visible to voters and journalists overnight, the two databases were not yet merged. The headline number was Spears in third place. Without the poll tapes, the technical artifact would have been invisible.

17.5 — The Investigation, the Hand Count, and the Certification Battle

DeKalb VRE acknowledged the issue quickly. By the morning of May 25, the Secretary of State’s office and DeKalb VRE had identified that the District 2 race showed a discrepancy and that a comparison of Election Day machine counts to expected results indicated something was wrong. Concurrent audits of other races on the May 24 ballot found no comparable anomalies. The error appeared isolated to District 2.

DeKalb VRE first attempted a machine recount — running the ballots through the scanners again. This produced unstable numbers and did not resolve the discrepancy, which is what would be expected if the underlying error was a mismatch between the BMD output and the scanner configuration: a second machine pass produces the same incorrect result, since the scanners are still misreading the printed selections.

A full hand count of all 40 District 2 precincts began on May 28. The hand count was conducted at DeKalb VRE’s offices and was open to candidates and the public. It ran over Memorial Day weekend, with a preliminary report released on June 1 and a certified count completed on June 3.

Certification did not come easily. The DeKalb Board of Registration and Elections missed its statutory certification deadline, voting unanimously at a May 31 special meeting to delay. Executive Director Keisha Smith told the board: “I do want to report that we are not done identifying an accurate tabulation. We’ve worked diligently, throughout the night, but I cannot provide an accurate calculation and tabulation of votes. Thank you for supporting us through this effort, but I’m not confident what I have is what I want to present to this board for certification.” The board reconvened on June 3, three days past the deadline, and voted 4-to-1 to certify the hand count.

In the certification meeting, several requests for an additional official recount were denied. Marshall Orson, who had been displaced from the runoff by the hand count, requested that the Georgia Bureau of Investigation open a criminal investigation. The GBI declined. Director Vic Reynolds cited two reasons: the relevant statute had not yet taken effect (it did not become operative until July 1), and Orson had not alleged any specific criminal conduct. State Representative Mary Margaret Oliver (D-Decatur) publicly supported Orson’s request, but the declination stood.

The certified hand-count totals, with the corresponding initial machine-count totals for comparison:

	Machine count May 24	Certified hand count June 3	Difference
Lauren Alexander	4,866 (34.6%)	4,878 (30.7%)	+12
Donald Broussard	0	135 (0.9%)	+135
Marshall Orson	5,527 (39.3%)	4,032 (25.4%)	−1,495
Michelle Long Spears	3,670 (26.1%)	6,823 (43.0%)	+3,153
Total Votes	14,063	15,868	+1,805
Blank Votes	4,607	2,779	−1,828
Times Cast	18,670	18,647	−23

The total ballots cast — Times Cast — differed by only 23 (DeKalb forensic analysis; AJC) between the machine count and the hand count. The total turnout was nearly identical. What changed was the attribution of votes: roughly 1,800 ballots that the scanners had recorded as “blank” in the District 2 race actually contained votes for candidates, and roughly 1,500 votes that had been credited to Marshall Orson actually belonged to Spears or to the blank column. Lauren Alexander, who appeared first on the ballot and was therefore unaffected by the shift, had her totals confirmed almost exactly.

The June 21 runoff, which DeKalb VRE had originally described as Orson vs. Alexander, became Spears vs. Alexander. Spears won the runoff with 61% of the vote. She took office as DeKalb County Commissioner for District 2 in January 2023.

17.6 — The Failed Checks

Every official safeguard prior to certification either missed the error or could not have caught it.

Logic and accuracy testing. Georgia conducts pre-election logic and accuracy (L&A) testing on all voting equipment, in which sample ballots are run through the system and the results compared to expected outcomes. DeKalb’s L&A testing in early May had not detected the configuration mismatch — for the simple reason that the mismatch was introduced after L&A testing, by the text-fix update pushed during early voting. L&A testing operated on the original configuration; the configuration that ran on Election Day was a different one.

Pre-certification audit. DeKalb’s normal pre-certification process, which compares aggregate totals across multiple data sources, did not catch the error during routine review. The error was caught only after Spears and her campaign physically visited precincts and gathered photographic evidence. State Elections Director Blake Evans’s June 3 statement framed the discovery as a vindication of Georgia’s system: “DeKalb County’s ability to quickly identify a discrepancy with their election results is evidence that Georgia’s election system works and is secure.” But the discrepancy was identified by a candidate, not by the system. DeKalb VRE’s role was to respond to the candidate’s evidence, not to detect the error independently.

The election summary report. The publicly posted overnight summary contained internal evidence of the error: the Database 1 report showed Spears at 23.98%, the Database 2 report showed her at 44.87%. A reader who knew to compare them could have seen the inconsistency. But the reports were posted as separate files, neither merged nor cross-referenced, and the public-facing top-line numbers did not flag the divergence.

Concurrent audits of other races. When DeKalb VRE confirmed that the issue was confined to District 2, it did so by running concurrent audits of other races on the May 24 ballot. This is a meaningful check against the alarming possibility that the configuration error had affected races beyond District 2. But it is a check that runs only after a problem has already been identified and only on the prompting of an unrelated discovery. It does not detect a problem that no one has yet noticed.

The pattern parallels other cases in this collection. In Monmouth County, New Jersey, in 2022 (see Case 16 — Monmouth County, New Jersey, 2022), the official post-election audit reported “100% tabulation accuracy” while the wrong candidate had taken office. In Butler County, Ohio, in 2008, official audits did not detect a software bug that had been silently dropping votes for nearly a decade. In Shelby County, Tennessee, in 2015 (see Case 9 — Shelby County, Tennessee, 2015), the system that should have caught a 40% precinct-level vote drop was a single photograph taken by a private citizen named Bennie Smith. In each of these cases — and in DeKalb in 2022 — the error was identified not by the system designed to catch it, but by an outside observer who happened to know enough to recognize that something was wrong.

17.7 — The DeKalb Analysis

In July 2022, six weeks after certification, DeKalb VRE released a detailed forensic analysis of the error. The eight-page document, titled “May 24 Commissioner D2 Analysis,” is exceptionally clear and rigorous. It identifies the four equipment configurations, classifies all 40 precincts, and tests the hypothesis that the ballot shift alone explains the entire discrepancy between the machine count and the hand count.

The methodology is straightforward. The DeKalb staff took the certified hand-count results — which they treated as the “correct” count — and applied the known shift transformations to each precinct’s numbers: at the 29 fully shifted precincts, they moved Spears votes to Orson and Orson votes to blank; at the four “Broussard touchscreen” precincts, they moved any Broussard votes to blank; at the four correct precincts, they applied no transformation; at the three mixed precincts, they applied the appropriate transformation per scanner. The result was a “predicted” machine count generated entirely from the hand count plus the known shift pattern.

The predicted machine count and the actual machine count match almost exactly. The total Election-Day vote was predicted at 7,536 and actually recorded at 7,546 — a difference of 10 votes across 32 affected precincts. The predicted Times Cast was 10,988 vs. an actual 11,001 — a difference of 13. Per-candidate predicted totals: Alexander 3,004 (matched exactly to hand count); Orson 3,509 vs. actual 3,524; Spears 1,023 vs. actual 1,029. The remaining differences are at the level of normal hand-count variance — the kind of small errors that occur whenever humans count ballots by hand.

The PDF’s conclusion: “There was a large error introduced into the machine count by the ballot shift; but we do not see any evidence of other sources of error.”

This is a strong claim, and it is supported by the forensic reconstruction. The ballot shift fully explains the discrepancy. It is not a partial explanation; it is the complete explanation, modulo small hand-count variance. There is no residual that would suggest additional unexplained errors, no precinct where the shift transformation fails to produce the observed machine count, no candidate whose totals don’t reconcile.

The DeKalb GOP’s request, around the same time, for a countywide hand recount of all races on the May 24 ballot — issued by Chair Marci McCarthy — was declined. The forensic analysis is the strongest counter to that request: if the ballot shift fully explains the District 2 error, and concurrent audits of other races showed no comparable anomaly, then there is no evidence basis for extending the recount further. But the request itself is instructive. In the absence of independent verification of other races, DeKalb VRE’s assurance that the error was confined to District 2 was an institutional claim that skeptics had no way to validate without performing the audit themselves. The lack of independent evidence — which is the recurring theme of this collection — is what created the demand. The pattern was visible elsewhere: the DeKalb episode prompted parallel challenges to results in Chatham County, even though no comparable error had been documented there. Once the public learns that an unverified reporting layer can produce a wrong winner once, the absence of independent verification anywhere becomes its own grievance.

17.8 — What Actual Vote Would Have Contributed

This case requires a careful assessment because it sits in an unusual position relative to Actual Vote’s standard methodology.

The standard AV pitch does not directly apply

Actual Vote’s core workflow compares poll tapes (printed at the precinct on election night) against officially reported results (aggregated by the election management system and published by the county). The comparison detects errors that occur between the precinct count and the official report — errors in transmission, aggregation, or final reporting. Errors that occur upstream of the poll tape — at the counting layer, where the ballot is read and the count is generated — typically produce poll tapes that already reflect the wrong count. The official result then matches the poll tape, and the AV comparison shows no discrepancy.

The DeKalb error is a counting-layer error. The precinct scanners read the ballots incorrectly because of the BMD-scanner configuration mismatch. The poll tapes printed by those scanners reflected the incorrect counts. When the data was uploaded to the county’s election management system, no further error was introduced — the official Database 1 report on May 25 showed essentially the same incorrect counts that had been printed on the poll tapes. A naive AV comparison of poll tape totals against the official county summary would have found the two matching.

But AV catches it anyway, by a different mechanism

The catch in this case comes not from the comparison between poll tape and reported result, but from the content of the poll tape itself. A poll tape reporting zero votes for a candidate at the precinct where that candidate voted is structurally impossible. A poll tape reporting zero votes for a candidate in 32 of 40 precincts where that same candidate received 6,800-plus votes in advance voting is nearly as impossible.

What AV provides — and what Spears used by hand — is the public, immediate availability of the poll tape data in a form that allows comparison against ground-truth knowledge. Spears knew she had voted for herself. She knew her husband had. She knew, from her campaign data, that her support in those precincts was nontrivial. The poll tapes were the formal evidence that contradicted what she knew. Without the tapes posted publicly, she would have had nothing to escalate. With them, she had everything.

If AV had been deployed for the May 24 primary, several things would have been true:

Aggregation across all 40 precincts within hours rather than overnight in person. Spears spent the evening of May 24 and the morning of May 25 driving precinct to precinct, photographing tapes by hand. AV’s volunteer infrastructure would have collected the same data automatically as polls closed. The pattern — Spears at zero in 32 precincts, at meaningful totals in 8 — would have been visible in aggregated form by the time the unofficial overnight numbers were posted, not three days later when the campaign assembled enough photos to bring to media.
Wider distribution of the evidence. The photos Spears collected were valuable but had to be physically circulated, brought to media in person, and presented at the May 31 special meeting before they prompted institutional response. AV would have made the precinct-level data visible to anyone with a browser: supporters, journalists, the Democratic Party (which Spears had to spend days lobbying before they backed her), the opposing campaigns, and election observers who could provide independent confirmation. The institutional escalation Spears had to perform manually would have been distributed across many more channels.
An independent record outside DeKalb VRE’s custody. All of the evidence Spears assembled was photographic; AV’s evidence is also photographic but is timestamped, cross-referenced, and held outside any government custody. In a hypothetical alternative timeline where DeKalb VRE had not acknowledged the issue, or where the institutional response had been slower or more contested, AV’s record would have constituted an independent evidence base for legal or political escalation.
The “For Candidates” use case made literal. Spears was, by her own statement, looking for early vote-total data for campaign purposes when she discovered the error. AV is designed precisely for that workflow: candidates and their teams using the system to monitor results in their own precincts, identify anomalies, and gather evidence in real time. The Spears case is the strongest single argument for why a candidate-facing AV experience matters: a candidate who knows their own ground truth is uniquely positioned to detect impossible poll-tape numbers, and the existence of the tool means that detection happens in hours rather than requiring a frantic cross-county drive.

What AV would not have done

AV would not have diagnosed the technical cause of the error. AV detects that the poll-tape numbers are inconsistent with knowable reality; it does not diagnose whether the cause is a BMD-scanner mismatch, a memory-card upload failure, a software defect, or something else. The forensic analysis that DeKalb VRE produced — identifying the four equipment configurations and reconstructing the ballot shift — is institutional work that AV does not replace.

AV would also not have produced the hand count itself. The ultimate corrective action — counting all 40 precincts by hand — required the institutional cooperation of DeKalb VRE, the Secretary of State’s office, the Board of Registration and Elections, and (during the certification dispute) the willingness to delay certification past the statutory deadline. AV provides evidence that triggers institutional response; it does not substitute for institutional response.

What AV provides is the trigger, and a trigger that arrives faster, with more redundancy, and outside the chain of custody whose work is being checked.

17.9 — The Counting-Versus-Reporting Question

The case study methodology document for this collection makes an explicit distinction between counting-layer errors (upstream of the poll tape) and reporting-layer errors (downstream of the poll tape). It notes that AV’s standard methodology operates at the reporting layer. The DeKalb case complicates the distinction in a useful way.

The error here was unambiguously at the counting layer in the sense that the wrong numbers were generated by the precinct scanners and printed onto the poll tapes. But the error was catchable via poll-tape transparency for two reasons that do not depend on the standard AV comparison:

First, the error was extreme. A position-shift bug that produces zero votes for a candidate in 32 of 40 precincts is not a small error. It is large enough that anyone with knowledge of the candidate’s actual support — the candidate herself most obviously, but also her family, supporters, campaign staff, opposing campaigns, journalists with access to early-voting numbers, and political observers — can recognize the impossibility on inspection.

Second, the cross-precinct pattern was visible. Spears had non-zero results in 8 precincts and exact-zero results in 32. Such a binary pattern is not a normal feature of voter behavior. It indicates either an extraordinary geographic clustering of opposition to the candidate (in this case, contradicted by her early-voting numbers) or a technical artifact affecting some precincts but not others. Pattern-recognition across the full set of poll tapes is one of the things AV’s aggregated view enables; even without the explicit AV comparison, the public availability of all 40 tapes makes this kind of anomaly detection possible.

The case therefore expands the practical envelope of what AV can flag without changing the methodology itself. The methodology comparing poll tapes to reported results catches reporting-layer errors. The same methodology, applied to enough poll tapes with enough public visibility, also catches counting-layer errors when those errors are large enough or patterned enough to be visible against ground-truth knowledge. This is an additive capability, not a replacement for institutional safeguards, but it is not nothing — and in DeKalb, it would have been the difference between an error caught in hours and an error caught after three days of personal driving by the affected candidate.

A useful way to put the distinction: AV does not promise to catch every counting-layer error. It promises to catch reporting-layer errors. But AV’s poll-tape transparency is a public good independent of the AV comparison itself, and in cases where counting-layer errors are externally falsifiable — visible against knowable reality — that transparency catches them too.

17.10 — The Vulnerability Beyond DeKalb

The DeKalb error was caused by a specific cascade: a candidate withdrawal, a redistricting fix, a BMD text fix, and a procedural mistake in which the third update was built from the wrong base copy. No single piece of the cascade is particularly unusual. Candidate withdrawals after the ballot is built are common; mid-cycle redistricting adjustments occur whenever census-driven district lines are finalized late; BMD configuration patches are routine. The danger is in the combination — and in the fact that the safeguard against the combination, logic and accuracy testing, was conducted before the final patch was applied.

The Secretary of State’s Center for Election Systems handles ballot programming for many Georgia counties, not only DeKalb. In each, the same general pattern applies: an election project is built; modifications are made in response to logistical requirements; updates are pushed to BMDs and scanners through memory cards. The DeKalb cascade was not specific to DeKalb’s hardware or software. It was a consequence of how the workflow handles late-stage modifications. Any Georgia jurisdiction that experiences a similar combination of mid-cycle changes — or any jurisdiction in any state running similarly architected election-project workflows — is potentially exposed to a comparable failure.

Georgia conducted statewide L&A testing again in advance of the November 2022 general election; the GovTech / Atlanta Journal-Constitution coverage of that testing cycle (September 2022) explicitly cited the DeKalb May 24 incident as motivation for stricter procedures, including more comprehensive testing of every position on every ballot style. State Elections Director Blake Evans’s June 3 statement framed the case as an example of the system working — meaning that the post-election review caught the error before certification. That framing is reasonable on its own terms but underweights the role that an outside party (the candidate herself) played in the discovery.

The deeper vulnerability is structural. Late-stage configuration modifications, when applied across complex election infrastructure with multiple distinct equipment types (BMDs, scanners, EMS servers), introduce opportunities for precisely the kind of mismatch that occurred in DeKalb. Pre-election L&A testing operates on the configuration at the time of testing; modifications applied between testing and Election Day are not necessarily re-tested at the same depth. The DeKalb analysis document does not propose specific reforms beyond identifying the cascade, but the implication is clear: any modification to the configuration after L&A testing should trigger a comparable level of pre-deployment verification.

17.11 — What We Know and What We Don’t

What we know:

The District 2 race on the May 24 DeKalb primary ballot was affected by a configuration mismatch between BMDs and scanners at 32 of 40 precincts. The mismatch produced a positional shift in which votes for Spears were recorded as votes for Orson, votes for Orson were recorded as votes for Broussard (and then converted to blank because Broussard had been disabled in the EMS), and votes for Alexander — who appeared first on the ballot — were unaffected. The total ballots cast in the race were nearly identical between the machine count and the hand count (a 23-vote difference); only the candidate attribution was wrong. The error was caused by a four-step cascade: a candidate withdrawal, a mid-cycle redistricting fix, a BMD text-display fix, and a procedural mistake in which the text fix was built from the wrong base copy. The hand count, conducted over three days, produced a result in which the ballot shift fully explained the discrepancy with no residual evidence of additional errors. Spears finished first with 6,823 votes (43.0%), qualifying for and winning the June 21 runoff.

The error was discovered when Spears noticed it and gathered evidence by visiting precincts and photographing poll tapes. DeKalb VRE acknowledged the issue, conducted the hand count, and ultimately certified the corrected results. The Georgia Bureau of Investigation declined a request to open a criminal investigation. No formal independent investigation in the manner of the Harvey investigation in Monmouth County, New Jersey, was conducted; DeKalb VRE itself produced the forensic analysis published July 13, 2022.

What we don’t know:

Whether logic and accuracy testing protocols across Georgia have been updated to require re-testing after any configuration modification — and if so, in what specific terms. Whether other Georgia counties using parallel election projects (for redistricting fixes or other reasons) have been audited to confirm no comparable cascade has occurred undetected. Whether Spears’s poll-tape photographs are publicly archived; she described collecting more than a dozen, but no comprehensive public release of the images appears to have been made. Whether the four “correctly configured” precincts in DeKalb’s analysis represent precincts that escaped the cascade by accident or by design — that is, whether the BMDs in those precincts simply did not receive the text-fix update, or whether some other factor preserved their original four-candidate configuration.

What we also don’t fully know is the counterfactual that motivates this case study: in how many DeKalb-style cascades has a comparably extreme error gone undetected because no candidate happened to perform the kind of in-person poll-tape verification that Spears performed? The Spears case was caught because the candidate herself was the natural detector. In a race where no candidate has the energy, resources, or geographical familiarity to visit the affected precincts personally on election night, the same class of error could persist through certification and into office.

17.12 — Why It Matters

The case for Actual Vote does not rest on this single case study, but the case study illustrates several of the collection’s recurring themes with particular sharpness.

Discovery by individual persistence. The DeKalb error was caught because one candidate refused to accept what the official summary said. Spears drove from precinct to precinct on a weeknight, photographed tapes, lobbied her party, brought evidence to election officials, and persisted through three days of delayed certification. The pattern matches Bennie Smith in Shelby County, James Bean in Monmouth County, Eric Olsen in Prince William County, Betty McGary in Butler County. In each case, the institutional safeguards that should have caught the error did not, and the error was caught only because an individual outside the institutional chain noticed and refused to be deflected. This is a fragile foundation for a democratic system: it depends on the right person caring enough about the right race in the right place. AV is designed precisely to systematize what Spears, Smith, Bean, Olsen, and McGary each did individually.

The “For Candidates” case made concrete. Spears stated explicitly that she was looking for early-vote total data for her campaign when she encountered the zeros. This is the canonical “For Candidates” workflow: candidates monitoring their own precinct-level numbers in real time. The DeKalb case demonstrates that this monitoring is not merely a campaign-management tool; it is sometimes the only mechanism by which catastrophic errors are caught. Any Actual Vote outreach to candidates and their staff should reference this case directly. The pitch is not abstract: “if you are a candidate, you may be the only person who can catch certain kinds of errors against your own campaign, and you deserve a tool that lets you do it without having to drive precinct to precinct yourself.”

The structural value of poll-tape transparency. Georgia’s requirement that poll tapes be posted publicly outside polling places on election night was the precondition for everything that followed. Without that transparency, Spears would have had no evidence to present, the Democratic Party would have had no basis for backing her escalation, and the institutional response would have been delayed or absent. States with weaker poll-tape access — those that require records requests, appointments, or institutional cooperation — would not have produced the same outcome on the same timeline. The DeKalb case is a positive argument for the Georgia transparency model and against models that gate access behind procedural friction.

The cascade pattern. The DeKalb error was not a single failure but a sequence of individually-reasonable steps whose interaction produced a catastrophic mismatch. This pattern recurs across the collection: in Monmouth County, an unrelated network IP-address conflict triggered a software reinstallation that disabled a double-counting safeguard; in Butler County, an antivirus scanner was blamed for what turned out to be a software bug that had existed for nearly a decade. Complex election infrastructure is full of points where individually-sensible decisions can interact badly. Independent verification — running the comparison from a different angle, with different evidence, outside the institutional chain — is the only reliable way to catch interaction failures that no single safeguard was designed to detect.

The race did not change a winner who held office for any meaningful period — Spears went on to win the runoff, and there was no interim period during which a wrongly-certified winner served (unlike Monmouth County, where the wrong candidate served for two months on a school board). But the runoff lineup itself was the wrong lineup until the hand count corrected it. Marshall Orson, who finished second on election night, was eliminated; Lauren Alexander, who finished second after the hand count, advanced. The runoff matchup that Democratic voters in District 2 saw on June 21 was the result of the certified hand count, not of the machine count. Without the candidate’s persistence, the runoff would have been Orson against Alexander, and the eventual winner of that hypothetical runoff is unknowable.

The deeper point is the one that recurs across the collection: in many of these cases, the question is not whether an error was definitely consequential, but whether the public has sufficient evidence to be confident the result was right. In DeKalb in May 2022, that evidence existed only because a candidate took her phone to her own precinct and saw a number that was impossible. The infrastructure to make that evidence available to everyone, automatically, on every election night, in every jurisdiction with public poll-tape access, is what Actual Vote provides.

17.13 — Further Reading