Inside the Investigation

1. Why this page exists

The 2022 Wake County Midterm report states the result in a line: of 189 first-pass differences, 79 resolved to matches with a plausible explanation. This page shows the work behind that line — the actual correspondence with the elections office, the county records we followed, and the checks we ran — because the process is as much the point as the result. It is a worked example of how America Counts turns an unexplained difference into an accounted-for one without ever guessing, and without wearing out the elections office’s patience.

Nothing here changes the report’s numbers. It documents how they were reached.

2. Where the 79 came from

Each Actual Vote analysis begins by comparing a value transcribed from a poll tape to the officially reported value at the same grain (county × precinct × voting method × contest × choice). A difference on that first pass is an initial non-match — not yet a discrepancy, just something to explain.

In Wake County, 1,424 comparisons produced 189 first-pass differences. Review sorted them:

• 27 were plausibly equivalent write-in situations — the tape’s raw catch-all write-in count and the official results’ canvass-resolved write-in lines are the same votes represented two ways. The write-in equivalence check closes the arithmetic exactly, so these are matches.
• 83 were values with no possible comparison — tape values from partially recorded two-machine early-voting sites, where the official total aggregates both machines and we recorded one. Not discrepancies; structurally not comparable.
• 79 did not fall to either check. They were concentrated in three election-day precincts and are the subject of this page.

The 79 began as likely attributable discrepancies — differences for which we held a benign working hypothesis but had not yet confirmed. That is a transient state. What follows is how it resolved.

3. The three signatures

The 79 are not a scatter. Each precinct shows one clean pattern:

A signature this regular is itself a clue: one-vote offsets in one direction, confined to one precinct, are the fingerprint of a specific canvass action, not of random error or of anything adversarial.

4. The ask: our letter to the Board of Elections

On July 29, 2024, we wrote to the Wake County Board of Elections. The letter is deliberately specific — it names the precincts, the contests, the exact vote counts, and asks a factual question rather than making an accusation. This is the lighter ask that a likely-attributable difference warrants: we hold a hypothesis and simply want it confirmed. The five questions:

Question 1: In Precinct 17-04 election day voting, why does US Senate candidate Matthew Hoh have 15 votes on the poll tape but 16 votes in the official results? (Other contests in this precinct also have one choice in which the poll tape total is one less than the official total in the same way.)

Question 2: In Precinct 19-12 election day voting, why does US Senate candidate Cheri Beasley have 765 votes on the poll tape but 764 votes in the official results? (Other contests in this precinct also have one choice where the poll tape total is one greater than the official total in the same way.)

Question 3: In Precinct 19-14 election day voting, why does County Sheriff candidate Willie Rowe get 308 votes on the poll tape, but only 306 votes in the official results?

Question 4: In Precinct 19-14 election day voting, why do 27 contests have one choice in which the poll tape total is one greater than the official results (just like in Question 2)?

Question 5: Can you confirm how many tabulating machines would have been present at One-Stop sites OSCH, OSHT, OSJB, OSOP, and OSTA? … in these sites, I suspect that I have received some, but not all of the required poll tapes since my totals are systematically way-short.

5. The reply

The Board replied the next morning, July 30, 2024. Its answer names the canvass mechanisms that produce exactly these signatures:

Election night results (i.e. poll tapes) are unofficial until the Board completes the reconciliation process (Canvass) outlined in NC Law. For a period of 10 days after the election, our staff researches provisional ballots and addresses any other outstanding items. Write-ins must also be manually added if voter wrote in a candidate that was already listed. Tabulator jams can also occur at polling sites, causing the unofficial count to be different. If a provisional ballot is accidentally scanned on Election Day, that ballot would be retrieved (subtracted from results) and presented to the Board with the other provisionals. Depending on the situation and staff findings, the Board may decide to rescan a precinct or make another determination that would explain those differences. Our Board meeting minutes from 11/17/2022 and 11/18/2022 may have more details about these precincts. Regarding question 5, the list of DS200 tabulators deployed for the 11/8/2022 can be found in the attached PDF.

Three mechanisms, mapping onto the three signatures:

• Manual write-in addition → official +1 for a listed candidate (17-04).
• Provisional-ballot retrieval → official −1 where an election-day-scanned provisional was later removed (19-12).
• Tabulator jam and re-scan → the mixed pattern where duplicate scans are removed (19-14).

The reply also attached the DS200 certification list, which resolved Question 5 by confirming each one-stop site’s machine count.

A reply like this — complete, substantive, and next-day — is the cooperative-inquiry approach working as intended. A specific, evidence-backed, politely batched question got a specific answer.

6. Following the pointer: the canvass minutes

The reply pointed us to the Board’s canvass minutes of November 17–18, 2022. We obtained them. They bear the explanation out — unevenly across the three precincts, which is worth stating precisely rather than glossing.

Precinct 19-14 — named directly. The 11/17/2022 minutes record the issue and the fix by precinct number:

Director Gary Sims informed the Board of an issue at precinct 19-14 involving a tabulator jam and duplicate ballot scans and asked to rescan precinct 19-14 and accept the results. After discussion … the Board unanimously agreed to rescan 19-14.

and, later in the same meeting, the reconciliation is formally approved:

Director Gary Sims presenting the Board with reconciliations for precincts 02-06, 17-04, 19-14, and the Avery Street Early Voting location. … the Board unanimously agreed to approve precinct reconciliations of 02-06, 17-04, 19-14, and the Avery Street Early Voting location.

Removing duplicate-scanned ballots lowers the official total below the election-night tape — which is exactly the 19-14 pattern, including the Sheriff −2 (two duplicated ballots that both marked one candidate). This is no longer “consistent with a mechanism the county described”; it is the county’s own contemporaneous public record documenting the cause, the corrective action, and the Board’s approval.

Precinct 17-04 — named in the reconciliations. 17-04 appears in that same approved list, so the minutes confirm it underwent a formal, Board-approved reconciliation. They do not spell out its mechanism; the write-in-addition reading (official +1) still rests on the Board’s email. The direction is consistent: adding a write-in vote to a listed candidate raises the official total, which a de-duplicating rescan could not do.

Precinct 19-12 — mechanism documented, precinct not named. 19-12 is not in the reconciliations list. But the minutes do record the very mechanism cited for it occurring in this canvass:

… the Board unanimously agreed to deny one provisional ballot due to the voter not being registered, and to redact the vote since the provisional ballot was placed in the tabulator on Election Day.

A provisional scanned election-night, then subtracted, produces the official-one-lower signature exactly — but the minutes do not tie it to 19-12. This is the most inferential of the three, and we say so.

(For context, the minutes also note the sample hand-eye count came back “an exact match,” and the full canvass was approved and certified on the 18th.)

7. A second exchange: the Avery Street early-voting site

A separate exchange, months later, closes a loose end. In November 2024 a volunteer’s poll tape from the “EVBOE” early-voting site (early voting at the Board of Elections) showed the same kind of mixed off-by-one/two pattern. We asked, and the Board answered:

The site reported a couple of ballot jams throughout the voting period, and they were unsure if they ran the ballots through the machine again (essentially double counting the same ballot). To reconcile and confirm we were reporting the official totals correctly, we re-scanned all ballots for that site and updated the official election results for this site at canvass. This was done during a board meeting that was open to the public.

That “board meeting that was open to the public” is the 11/17 meeting above — the “Avery Street Early Voting location” rescan in the reconciliations list. So the EVBOE site is Avery Street, and its re-scan is documented in the same minutes. (This site is not one of the 79; it is an early-voting site, reported separately. It matters here only as an independent, worked confirmation that the re-scan mechanism is real and routine at WCBE — the mechanism behind 19-14.)

8. The mechanism-fit check

A plausible explanation should not just point in the right direction — the size of each difference should be within what the cited mechanism can actually produce. A provisional retrieval predicts a −1 on each contest the provisional voter marked; a rescan removing k duplicate ballots predicts −1 or −2 on the choices those ballots marked; a manual write-in addition predicts +1 for one listed candidate. We checked every one of the 79 against its mechanism’s predicted magnitude.

77 of 79 fall exactly within the predicted magnitude. The two that do not are both write-in lines, not candidate totals:

• 19-14 Raleigh Mayor, write-in: tape 176 vs official 15 (−161).
• 19-12 Soil & Water District Supervisor, write-in: tape 46 vs official 40 (−6).

A de-duplicating rescan cannot move a write-in count by 161, so mechanism-fit correctly ejects these two from the rescan/provisional buckets. They are the same phenomenon as the 27 write-in matches in Section 2 — the tape’s raw catch-all count versus the canvass-resolved write-in lines — in contests where the five-condition equivalence check did not fully close (for instance, because the tape’s own total line was not transcribed). They are write-in representation differences, not ballot-count changes, and they too are non-contest-flipping (write-in lines do not determine any candidate’s standing). Mechanism-fit does not change the disposition of any of the 79; it makes the per-value reasoning honest.

9. Magnitude and contest-flipping — the analyst’s judgment, recorded

Two conditions gate whether a plausibly explained difference is accepted as a match rather than pursued for direct confirmation: it must be non-contest-flipping, and of a magnitude the analyst judges benign. The second is a subjective call — there is no hard-and-fast rule for it — so America Counts records it, per precinct, so a reader can weigh (or dispute) the judgment:

Had any of these been large enough to matter to a contest’s outcome — contest-flipping — a plausible explanation would not have been enough; we would have pressed for direct, per-value confirmation. None was close. And had a non-contest-flipping difference been large enough that the analyst was simply not comfortable waving it through, that too would have triggered a more specific ask. The judgment is proportional, and it is on the record here.

10. Disposition: matches, with the door open

Each of the 79 has a plausible, non-contest-flipping, benign-magnitude explanation. We therefore resolve them to matches with a plausible explanation — a match, never a discrepancy, in the report’s statistics. The explanation is credible; for the 29 in 19-14 it is documented in the county’s own minutes; for the rest it rests on the county’s stated mechanisms and a clean mechanism-fit. We did not confirm every individual value’s arithmetic against the underlying canvass records, and we do not claim to have.

This is the same proportionality that governs how America Counts spot-checks transcriptions rather than re-verifying every one: when a low-stakes result is completely plausible under the documented circumstances, chasing 100% certainty is not a good use of our time or the elections office’s — and going round and round to fully resolve a one- or two-vote, non-contest-flipping difference would spend a finite, valuable relationship with the office for no change in any conclusion. What matters is that the reasoning is disclosed and the item stays open: every one of the 79 remains itemized, with its numbers and its explanation, and can be reopened at any time — by us, if new evidence arrives, or by any reader, who has all the poll-tape and official data needed to raise it with the office directly. All 79 are published in full, row by row, in a dedicated exhibit: The 79, in full →.

11. What this illustrates

The value of the Wake County analysis is not that it confirmed an outcome — at ~4.6% coverage it could not. It is that it demonstrates the whole funnel, end to end, on real data: record the tapes, transcribe them, compare, review, form a hypothesis, ask the office a specific and respectful question, follow the office’s pointer into the public record, check the mechanism against the magnitude, record the judgment, and disclose all of it. That is what an accountable, non-adversarial, independent check on vote reporting looks like — and it is the design this analysis was built to make routine in future elections, before certification, when a real discrepancy could still be corrected.

The correspondence and canvass records excerpted above are held in America Counts’ investigation file for this analysis. The underlying poll-tape recordings are public in the Actual Vote Video Archive. Return to the full report →