What the Voter-Roll Documents Establish — and What They Only Assert

Sports analytics taught me a rule I now apply to every document that crosses my desk: the box score and the postgame narrative are different things. So I've been running the most-circulated voter-roll records through one discipline — what does each document establish on its own face, and what does it only assert? On this subject the gap runs in both directions, and most coverage only reports the direction it likes.

The 278,000 number is really two numbers stapled together

Start with the figure driving the national story: a Department of Homeland Security review, cited on the White House's election-integrity page, that found roughly 278,000 noncitizens registered on federal voter rolls. The first thing to understand is that it isn't one finding. It's two, combined.

As the Spokesman-Review documented in breaking the DHS sheet apart, about 28,000 come from the two dozen or so states that ran their rolls through DHS's SAVE system — the federal Systematic Alien Verification for Entitlements database built to check immigration status. The other ~250,000 come from four states — California, Pennsylvania, New Jersey, and Nevada — that the DHS sheet itself says did not use SAVE, drawn instead from a review of "public voter files." Two methods, two very different confidence levels, and the released sheet doesn't explain either one or say why they were merged into a single headline.

That missing methodology is the legitimate criticism. David Becker of the Center for Election Innovation and Research called the combined figure "an irresponsible number" produced with "shoddy methodology," and on the narrow question of disclosure he's right: a government that wants a number believed should show how it got there.

But — and I'm correcting an earlier version of my own thinking here — a disclosure problem is not proof the underlying phenomenon is invented. The SAVE-based 28,000 aren't an estimate; they're records that hit against a federal immigration database. And where the mechanism is people who identified themselves as noncitizens, the resulting count is a floor, not an inflated ceiling — everyone who both wasn't a citizen and said so and got caught. The people who didn't say so, or weren't caught, aren't in it. To see why that flips the usual "debunking," look at the one state that just showed its work.

New Jersey: a Democratic governor confirms the mechanism is real

On July 21, 2026, New Jersey Governor Mikie Sherrill — a Democrat — announced that about 6,600 noncitizens had been registered to vote in her state between June 2023 and June 2024. This is not a partisan press release or a self-published tally. It is the state's own admission, delivered by the party with the least incentive to make it.

The mechanism is the part that matters. These weren't people an algorithm guessed about from a public file. They were people who told the state they were not citizens — and were registered anyway. As Sherrill described it, the system "refused to take no for an answer": even when applicants "checked 'No' on the keypad at the motor vehicle bureau," a software error in the Motor Vehicle Commission's system "still registered them to vote." Her words on the failure itself were not hedged — "I am appalled by the reckless failures that allowed this to happen," and "This failure didn't occur under my watch, but accountability starts now."

That is exactly the category I'd previously waved off as an inflated estimate. When the record is built from self-identification — a person affirmatively saying "I am not a citizen" — the number is the opposite of inflated. New Jersey is a working model of how the ~250,000 in the non-SAVE states could be real rather than imaginary, produced by the same motor-voter plumbing.

And then the discipline reasserts itself, because it also cuts the other way: of those 6,600 registrations, New Jersey's preliminary review found fewer than 400 actually cast a ballot. Registration is not voting. Six thousand improper registrations is a real, documented failure of the motor-voter system; it is not six thousand fraudulent votes. Both sentences are true, and a reader is owed both.

Colorado: the safeguard is real on paper, and it has already failed once

Colorado is where the structure gets uncomfortable, and where I want to separate statute from suspicion. Under SB 13-251, the Colorado Road and Community Safety Act, the state issues driver's licenses to residents regardless of immigration status — roughly 200,000 issued as of 2022. On paper the safeguard against those license-holders being registered is concrete, and it doesn't depend on any official's assurance: every SB-251 credential carries a printed black banner reading "Not Valid For Federal Identification, Voting or Public Benefit Purposes," and the motor-voter system is supposed to require proof of citizenship before it registers anyone. That marking is a real, checkable feature of the card, not a talking point.

Here's the problem with resting there. In October 2022, Colorado's Secretary of State mailed voter-registration notices to about 30,000 noncitizens — precisely the SB-251 license-holders the marking is supposed to filter out. The cause was a formatting error: the DMV list the state checked against arrived without the field that would have flagged and removed them. Secretary of State Jena Griswold's office said none of the recipients would actually be allowed to register if they tried. I'll report that she said it — but I won't launder an assertion by an elected official into a settled fact, and neither should anyone else. New Jersey just demonstrated that this exact category of assurance — the system won't let it happen — can be false in practice.

Be precise about what the 2022 event was and wasn't: it was 30,000 registration invitations sent in error, not 30,000 completed registrations, and not ballots. But the structural point survives that precision, and it's the one worth sitting with. Colorado mails a ballot to every registered voter automatically. That makes the registration filter the single point of failure between a noncitizen license-holder and a ballot in the mail — the same filter that just failed at scale in New Jersey, and that misfired once already in Colorado.

Now the other half of the ledger, because the standard doesn't get to skip it. The record of noncitizens actually completing registration and having votes counted in Colorado is thin. As the Colorado Sun noted in examining the question, the Heritage Foundation's election-fraud database — compiled by a conservative group with every motivation to find such cases — documents only two instances of noncitizens found guilty of attempting to register in Colorado across more than four decades. Confirmed, counted noncitizen votes are not established in the public record I can find. So the honest statement is narrow and it holds: the failure mechanism is real and proven elsewhere; a mass of counted Colorado votes is asserted, not documented. The question that would close the gap is one the state can answer and, as far as I can tell, hasn't published: run the current rolls against the SB-251 file and release the number. If it's near zero, that's the strongest possible defense of the system.

Leads, findings, and assertions

Not everything in circulation clears even the New Jersey bar. A widely shared volunteer review of Detroit's 2020 absentee records, for instance, concludes from a partial FOIA production that many ballots were counted without the legally required signed envelope — but the underlying dataset, the court order, and the county's own records aren't reproduced for a reader to check, and missing paperwork years later is not the same thing as fraud at the time. That's a lead worth chasing, not a finding worth citing. On the enforcement side, Just the News reports specific, dockets-attached prosecutions of noncitizen voting — the strong, checkable part — alongside softer aggregate figures attributed to DOJ officials rather than a produced case list. Named cases are findings; the aggregate is an assertion until the list appears.

Why this matters

Every one of these documents contains something real, and the honest work is refusing to round either direction. The national 278,000 figure has a genuine disclosure problem — two methods combined without showing the math — and that's worth saying plainly. But "the methodology sheet is sloppy" is not "noncitizens aren't on the rolls," and New Jersey just proved the difference with its own admission: the mechanism is real, it runs on self-identification the system overrode, and that kind of count is a floor. The same discipline then holds the other end — 6,600 registered, fewer than 400 voted, and no established mass of counted noncitizen votes in Colorado. Documented, inferred, asserted: I'd tag a CISA press release and a Media Matters rebuttal with the same three labels. The records that would settle the rest — Colorado's current count, the DHS methodology, the state-by-state case lists — mostly exist. The fix isn't louder adjectives in either direction. Produce the exhibits.