Case 15 — NC Supreme Court, 2020

15.1 — Introduction

This case study is different from the others in our collection. There is no undiscovered error, no evidence destruction, no wrong candidate seated. The system worked. Four tallies confirmed the same winner. Neither candidate alleged fraud.

And yet: the race for North Carolina Supreme Court Chief Justice in 2020 was decided by 401 votes out of nearly 5.4 million cast — a margin of 0.008%. Four different tallies produced four different margins of victory. At least two counties had reporting errors that were caught during the canvass but could easily have been missed. And the existing safeguards — a full machine recount, a sample hand-to-eye recount, and protests filed across 85 counties — still left open the question of whether every precinct’s results were faithfully reported.

This is not a story about failure. It is a story about what “good enough” looks like when the margin is 401 votes — and whether independent, precinct-level verification could have added the certainty that recounts alone cannot provide.

15.2 — The Race and Its Stakes

The 2020 contest for North Carolina Supreme Court Chief Justice pitted Democrat Cheri Beasley, the incumbent Chief Justice, against Republican Paul Newby, a longtime Associate Justice on the same court. Both were sitting members of the court they sought to lead.

The stakes were significant. The Chief Justice presides over the state’s entire court system — appointing commission members, selecting three-judge panels for sensitive cases involving wrongful convictions and redistricting, designating the chief judge of the Court of Appeals, and directing the Administrative Office of the Courts. During 2020, Chief Justice Beasley had used emergency directives to manage court operations during the pandemic, including suspending jury trials for most of the year.

The ideological balance of the court hung in the outcome. If Beasley won, Democrats would hold a 5-2 advantage. If Newby won, the advantage would narrow to 4-3 — and with two seats up for election in 2022, Republicans would have a plausible path to flipping the court. (They ultimately did so in 2022.)

North Carolina Republicans had created a judicial victory fund to support candidates across the state’s court system, and 2020 was a sweep year: the GOP won all eight statewide judicial races. The Chief Justice contest was the only one that was close.

15.3 — Election Night and the Canvass

On election night, November 3, 2020, the race was too close to call. Newby led by a small margin as election-day results came in, but hundreds of thousands of absentee-by-mail and provisional ballots remained to be counted.

This was the first North Carolina general election in decades in which every voter cast a paper ballot — the state had completed its transition to optical-scan tabulators with hand-marked paper ballots across all 100 counties. The scale of the count was enormous: 2,662 precincts, plus early voting sites treated as separate precincts for reporting purposes.

Over the following ten days, county boards of elections conducted their canvasses — processing absentee-by-mail ballots, adjudicating provisional ballots, and certifying results. The lead changed hands multiple times. By Thursday, November 12, as counting of late-arriving ballots continued, Beasley had pulled ahead. During canvassing on Friday the 13th, the lead shifted again as different counties reported at different times. By the end of the canvass day, Beasley held a lead of just 35 votes — with 1,472 provisional ballots from Robeson County still to be added.

Two county-level errors surfaced during the canvass:

Washington County had been using an outdated tabulator that inadvertently duplicated mail-in ballot data in its election night results, creating two records for every by-mail vote. The error was caught during required post-election audits and corrected during the canvass. When Washington County’s correction was applied, Newby retook the lead.

Robeson County discovered that early-voting ballots had inadvertently not been uploaded on election night. These were added during the canvass.

Both errors were caught through standard procedures. Neither required outside intervention. But both illustrate the fragility of the reporting chain in a race decided by hundreds of votes: a single county’s tabulator error was large enough to swing the statewide lead.

When the canvass was complete on November 17, Newby led Beasley by 406 votes. Beasley requested a statewide machine recount.

15.4 — The Machine Recount

Under North Carolina law, a statewide recount is mandatory when the margin is fewer than 10,000 votes. All 100 county boards of elections re-ran every ballot through a tabulator — nearly 5.4 million ballots in all.

The process was labor-intensive. Bipartisan two-person teams handled each tabulator: one person feeding ballots, one observing. Large counties like Wake (with approximately 635,000 ballots and an estimated cost of $110,000) took five or six days. Smaller counties like Greene (fewer than 9,000 ballots, labor cost of$ 400-$600) finished in a day. The entire process was completed by December 2, with Forsyth County crossing the finish line last — a week behind the original November 25 deadline.

The NCSBE explained that minor differences between canvassed results and recount results are expected: “Different machines may read stray marks differently.” The 2016 statewide recount in the state auditor’s race had produced a shift of about 50 votes out of 4.5 million cast, and the NCSBE estimated that the typical variation was roughly a couple of votes per 100,000 ballots cast.

The county-by-county recount data shows the texture of these small variations. Most counties shifted by zero to five votes for each candidate. A few showed larger movements:

Forsyth County: Beasley +30, Newby -26 — the largest single-county swing, a net shift of 56 votes toward Beasley
Catawba County: Beasley -21, Newby -22 — both candidates lost roughly equal numbers of votes
Rutherford County: Beasley +6, Newby +13

Several counties had recanvassed between the initial canvass and the start of the recount, bringing the pre-recount margin to 416 votes. After the machine recount, the margin was 401 votes — a net shift of 15 votes toward Beasley.

The recount confirmed the same winner. But it produced a different number.

15.5 — The Protests

While the recount was underway, both campaigns were simultaneously pursuing election protests — a separate legal mechanism under North Carolina law for challenging specific ballot adjudication decisions.

Newby’s protests came first. His campaign filed in eight Democratic-leaning counties — Duplin, Durham, Guilford, Mecklenburg, New Hanover, Robeson, Scotland, and Wake — alleging that county boards had accepted absentee-by-mail ballots that were not postmarked by Election Day, were not received on time, or had incomplete return-envelope information. These protests grew out of a broader legal fight: a September 2020 settlement between the State Board of Elections and a Democratic-aligned group had changed the rules for which by-mail ballots could be counted, and Republicans had unsuccessfully challenged the settlement in both state and federal courts, including two petitions to the U.S. Supreme Court. All eight counties dismissed Newby’s first-round protests. He filed a second round focused on ballots cast by voters who had died before Election Day — ballots that are invalidated by North Carolina law.

Beasley’s protests were filed in 85 counties on November 17, the same day she requested the recount. Her campaign alleged that over 2,000 absentee and provisional ballots had been wrongfully rejected across the state. In her letter to the State Board, Beasley did not allege fraud or mismanagement. She pointed to the complex and changing legal landscape for mail-in ballots in 2020: “Even in a year without the twists and turns we have seen in 2020, county boards make some errors. However, this year presented unique challenges to the candidates, the voters and especially our 100 county boards.”

The protest results were modest. County boards either dismissed the protests outright or approved small numbers of ballots — Brunswick County approved one provisional ballot; Durham County approved three. By the time Carolina Public Press reported on the process in early December, the protests had yielded approximately 14 additional votes across the state. This was not nearly enough to close a 401-vote gap.

Both candidates ultimately withdrew all protests after the hand-to-eye recount was completed.

Neither campaign alleged fraud.

15.6 — The Hand-to-Eye Sample Recount

Within 24 hours of the machine recount’s completion on December 2, Beasley requested the next available step: a hand-to-eye recount of a random sample of precincts, as provided by N.C.G.S. § 163-182.7A.

The procedure was detailed and carefully structured. Each of the 100 county boards selected 3% of its precincts by random drawing (conducted publicly by the State Board on December 4), rounded up to the next whole number. Early voting sites counted as separate precincts. Bipartisan four-person teams conducted the count at each site: two officials (one Democrat, one Republican) to read each ballot aloud, and two officials (one from each party) to separately record tallies on paper using a five-mark tally system, calling out “tally” at every fifth vote. If the two talliers did not call “tally” at the same time, the team had to resolve the difference before proceeding. County board members supervised but could not participate on counting teams.

The theory behind the hand-to-eye recount is that human counters can detect voter intent that machines miss — voters who circled the oval instead of filling it in, for example, or who placed a mark slightly outside the target area. The NCSBE issued a 10-page guidance memo with detailed illustrations of which marks should and should not be counted.

But the hand-to-eye recount also introduces its own sources of variation. The mandatory post-election audit of the presidential race — 292,158 ballots counted by hand — had found 42 total discrepancies between machine and hand counts. Most were attributed to errors by the human counters. Only 17 involved voter intent that the machine had missed.

The hand-to-eye sample recount ran from December 7 through December 15. In the sampled precincts, Beasley gained 28 votes and Newby gained 39 — a net shift of 11 votes toward Newby. Added to the machine recount margin, this produced a running figure of Newby +412, though it was the machine recount’s 401 that the NCSBE eventually certified.

The results did not come close to the threshold that would trigger a full statewide hand-to-eye recount. Under state law, such a recount would be ordered only if extrapolating the sample’s results to the entire state would reverse the outcome.

On December 12, as the hand-to-eye recount was nearing completion but not yet finished, Beasley conceded. “Today, I called Justice Newby to congratulate him on winning the election for Chief Justice of the North Carolina Supreme Court,” she said. “I offer my very best to him and his family as he moves into that new role.”

15.7 — Certification

On December 18, the State Board of Elections voted unanimously to certify Newby’s victory. Both candidates had withdrawn their protests. All recounts were complete.

“Thank you to the county boards of elections that worked day in and day out to ensure accurate results for North Carolina voters,” said Karen Brinson Bell, the State Board’s executive director. “The results for every contest under state jurisdiction in the 2020 general election are now certified.”

Paul Newby was sworn in as North Carolina’s 30th Chief Justice at midnight on January 1, 2021.

The official certified totals: Newby 2,695,951; Beasley 2,695,550. Margin: 401 votes.

15.8 — The Arithmetic of Uncertainty

Four different tallies of the same ballots produced four different margins of victory:

Count	Margin	Notes
Initial canvass (Nov 17)	Newby +406	After county corrections
Post-recanvass (before recount)	Newby +416	Several counties recanvassed
Machine recount (Dec 2)	Newby +401	All ~5.4M ballots re-tabulated
Hand-to-eye sample adjusted margin (Dec 15)	Newby +412	Machine recount margin plus the +11-vote net shift in the 3% sampled precincts. Not a statistical extrapolation to the full state; NCSBE certified the machine recount margin of 401.

The range across these counts is 15 votes. All four confirmed the same winner. But for a race decided by 401 votes out of 5.4 million, even a 15-vote range invites the question: where exactly does the uncertainty live?

The NCSBE’s own explanation is instructive: stray marks, partial erasures, and marginal fills may be read differently by different machines, producing “roughly a couple of votes per 100,000 ballots cast.” Scale that to 5.4 million ballots and you get an expected variation of approximately 100 votes — smaller than the margin, but not negligibly so.

Then there are the sources of variation that machines don’t explain: the Washington County duplication, the Robeson County upload failure, the Halifax County recount where 98 ballots were temporarily misplaced (they were found stored with authorization-to-vote forms). Each of these was caught and corrected. But a race decided by 401 votes across 100 counties and 2,662 precincts has 2,662 opportunities for small errors to go unnoticed.

And the existing checks don’t cover every precinct. The mandatory post-election audit covers two randomly selected precincts per county — 200 out of 2,662, or about 7.5%. The hand-to-eye sample recount covers 3% of precincts per county. The machine recount covers all precincts, but compares the same ballots against different machines — it tests counting accuracy, not reporting accuracy.

None of these checks systematically compares what was printed on each precinct’s poll tape against what was officially reported. In this respect, the NC Supreme Court race shares a structural feature with Florida 2000 (see Case 4 — The 2000 Presidential Election in Florida): when the margin approaches the system’s inherent noise floor, the distinction between “we got the same winner” and “we are confident in the margin” becomes critical — and that confidence depends on checks that cover the entire reporting chain, not just the counting layer.

15.9 — What AV Would Have Contributed

North Carolina’s election infrastructure is, in many respects, well-suited for independent verification. The state requires poll tapes to be printed and posted at precincts on election night. Public photography of posted tapes is explicitly permitted. Precinct-level results are published online and updated every five to ten minutes on election night. Historical precinct-sorted certified results are available for the past twenty years.

This means Actual Vote could have operated in North Carolina in 2020 — and what it would have contributed is distinct from what any of the existing checks provided.

The Washington County error. This was a reporting-layer failure: the tabulators counted correctly and printed correct tapes, but the election night results showed inflated totals because mail-in ballot data was duplicated. An AV comparison of tape totals against reported results would have flagged this discrepancy on election night — before the canvass, before the recount, before the lead changed hands. The error was caught through standard procedures, but AV would have caught it independently, immediately, and with evidence held outside government custody.

The precinct-by-precinct map. The machine recount produced county-level change data showing that Forsyth County alone shifted 56 votes net toward Beasley, while most counties shifted by zero to five votes. But the recount data is reported at the county level. An AV comparison would have produced precinct-level data — identifying exactly which precincts shifted and by how much, providing a granular map of where the variation concentrated. In a race decided by 401 votes, that granularity matters.

The reporting chain. The machine recount tests whether tabulators read ballots consistently. The hand-to-eye recount tests whether human counters agree with machines. Neither test systematically checks the step between them: whether precinct-level tabulator totals were faithfully transmitted to county totals and then to state totals. This is the step where Prince William County’s error occurred (see Case 12 — Prince William County, Virginia, 2020), and it is the step that AV is specifically designed to verify. In North Carolina, where precinct-level data passes through county aggregation before reaching the state dashboard, each aggregation step is a point where a transcription error, software bug, or upload failure could alter the reported total without affecting the underlying ballots.

The confidence argument. Four tallies confirmed the same winner. That is meaningful. But all of those counts were conducted by the same government entities using the same equipment, and all required the public to trust that the reporting chain between tabulator and official result was intact. AV provides a check from outside that chain — one that starts at the poll tape and is held by independent citizens. In a race decided by 0.008%, the difference between “the government counted again and again and got the same answer” and “the government counted repeatedly and an independent check confirmed the result” is not trivial.

What AV could not have addressed. AV is not relevant to the ballot-adjudication disputes that dominated the post-election legal proceedings. Whether a particular absentee ballot should or should not have been accepted, whether a voter who died before Election Day should have their ballot counted, whether the September settlement on mail-in ballot rules was properly applied — these are questions about which ballots to count, not about whether counted ballots were accurately reported. AV compares tape totals to official results. It has nothing to say about whether the tapes themselves reflect the correct set of ballots.

The defensive frame: close elections and the limits of institutional confirmation

The North Carolina 2020 Supreme Court race ended without either candidate alleging fraud, which is itself a meaningful fact and a credit to the candidates and the state’s election infrastructure. But the absence of an actual bad-faith dispute does not eliminate the structural vulnerability that produces such disputes; it only means the vulnerability did not become visible this time. Reading the case under the defensive frame highlights what a different ending would have required.

The closest statewide election in North Carolina history was resolved by four institutional counts inside a single custody chain, all of which produced the same winner within a fifteen-vote range. The institutional process worked. But every count was performed by the same government entities, using equipment from the same vendors, against the same source of original data, and arriving at conclusions that the public was asked to credit on the basis of “we counted again, and we got the same answer.” For a race decided by 401 votes out of 5.4 million, the institutional confirmation rests on a relatively narrow base: trust in the same institutions whose original count is being checked.

Now imagine the counterfactual in which one of the candidates had been less restrained, or in which the political environment had been more polarized, or in which a third party with no stake in the formal contest had decided to amplify a fraud narrative around the result. The available rebuttal to “the recounts were fixed” is “but they were rigorous, and the Republican-led Board of Canvassers and the Democratic Chief Justice both signed off.” That is true and matters. It is also the structurally weak answer, because it asks the accuser to trust the same institutions whose original conduct is being challenged. In a 0.008% margin race, the gap between “trust the institutions” and “look at the independent evidence” is exactly the gap a bad-faith narrative would exploit.

What AV evidence would have contributed in any such counterfactual is the missing non-institutional layer. Citizen-held photographs of precinct-level tape totals, captured at the close of polls on November 3, 2020, would have constituted a record of the underlying scanner output that any post-election dispute could have been tested against — independent of the State Board of Elections, independent of the county boards, independent of the candidates and their parties. A “the recounts were rigged” narrative would have to explain why thousands of citizen-held photographs of poll tapes happened to match the rigged result. A “the original count was wrong” narrative could be tested precinct by precinct against the photographs. The institutional process would still have done the institutional work. What AV adds is the layer of evidence that the institutional process cannot itself produce — the layer that close-election narratives most need when they arise.

The 2024 Riggs–Griffin North Carolina Supreme Court contest, which followed a similar pattern of an extremely close race resolved by institutional process, eventually generated the kind of post-election dispute that 2020 avoided. AV evidence in either race would have served the same defensive function. See Defending Against Bad-Faith Accusations for the broader frame.

15.10 — What We Know and What We Don’t

We know that four tallies confirmed the same winner, with a 15-vote range across those tallies. We know that neither candidate alleged fraud. We know that the existing safeguards caught at least two county-level errors during the canvass.

We know that the margin — 401 votes, or 0.008% — was the closest statewide election in North Carolina history. We know that FairVote’s analysis of statewide recounts from 2000 to 2019 found an average margin change of 430 votes — larger than Newby’s entire margin of victory. [JASON: The FairVote statistic is cited by Carolina Public Press from a FairVote report. Please verify you’re comfortable with this source and the specific number before publication.]

We don’t know whether other precincts had small reporting discrepancies that fell within normal variation and were never identified. The mandatory post-election audit covered approximately 7.5% of precincts. The hand-to-eye sample covered 3%. The machine recount re-tabulated all ballots but did not check the reporting chain. No systematic precinct-by-precinct comparison of tape totals to official results was conducted.

We don’t know the precise scope of the Washington County duplication error. The final certified results for Washington County show 654 absentee votes in the Chief Justice race (Beasley 460, Newby 194). If these were fully duplicated, the error would have added approximately 654 phantom votes with a net inflation of 266 votes toward Beasley — consistent with the fact that correcting the error swung the statewide lead from Beasley to Newby. But we do not have the pre-correction numbers to confirm the exact magnitude, and the details received limited public scrutiny. [JASON: Can you verify the duplication math from NCSBE precinct-level data? The pre-correction election night results for Washington County would confirm the exact impact.]

15.11 — Why It Matters

Most of our case studies are about elections where something went wrong and nobody caught it — or caught it too late. This one is about an election where the checks worked, the result was confirmed, and both candidates moved on.

So why include it?

Because “the checks worked” is not the same as “the checks were complete.” Four tallies and 85 counties’ worth of protests produced a confirmed result, but not a comprehensively verified one. The existing safeguards checked counting accuracy (machine recount), voter intent (hand-to-eye sample), and ballot eligibility (protests). None of them checked reporting accuracy — the step between the tabulator’s poll tape and the official result.

For most elections, this gap does not matter. When the margin is measured in thousands or tens of thousands of votes, small reporting discrepancies disappear into the noise. But when the margin is 401 votes out of 5.4 million — when a single county’s tabulator error can swing the statewide lead — the reporting layer becomes the margin of uncertainty.

Actual Vote is designed to close that gap. Not to replace recounts or audits, but to add the one check that neither provides: a precinct-by-precinct comparison of what the tabulators printed against what was officially reported, conducted independently and held outside government custody.

In this race, AV evidence that confirmed the official results would have been as valuable as evidence that contradicted them. Either outcome serves the same purpose: giving the public a basis for confidence that is independent of trust in any single institution.

For an election decided by 0.008%, that matters.

This is the quiet case for the transparency half of Actual Vote’s mission. In the loud cases — Baldwin County, Ohio 2004, LBJ 1948 — the absence of independent evidence is what makes a controversy permanent. In the quiet cases — the ones resembling this one, where nothing dramatic went wrong and both sides accepted the result — the absence of independent evidence is what leaves even a well-run, closely-contested election short of the evidentiary standard its stakes deserve. A race for Chief Justice of a state Supreme Court, decided by 401 votes across 2,600 precincts, through four separate tallies and eighty-five counties’ worth of protests, and finally conceded without fraud allegations, is close to the best that the current system produces. Even so, no independent precinct-level evidence exists against which any citizen of North Carolina can check the result. AV’s contribution to cases like this is not to expose hidden problems. It is to convert “we trust the process produced the right answer” into “we can show the process produced the right answer” — a different and stronger thing. That conversion is the transparency benefit, and it is the one a race this close most needed.

15.12 — Further Reading

Cross-references in this collection

Case 12 — Prince William County, Virginia, 2020 — A reporting-layer failure that AV is designed to catch and that statewide RLAs don’t.
Case 4 — Florida 2000 — Another close race where the reporting layer became the margin of uncertainty.