Define the System Before You Fund It

Congress can require proof of citizenship. It should not fund the system before deciding how verification and correction actually work.

For years, my job at the Small Business Administration (SBA) was to review eligibility after federal money had already gone out the door. I reviewed the loan file, applied the guidance, and determined whether the documents supported what the applicant had claimed. That work taught me the distinction: collecting a document is not the same as verifying eligibility.

I am not an election administrator or an election policy expert, and I will not pretend to be. What I know is verification. What it takes to make a determination that holds up, and what happens when a program gets funded before anyone decides how the determination gets made.

Congress is taking up that distinction on two tracks moving at different speeds. The House attached the SAVE America Act (the Act) to a national security appropriations bill, and then attached it again to the defense authorization bill, which passed 216-212 on July 22 under a rule directing the Clerk to add the Act’s text at engrossment. So, the House has now sent the requirement toward the Senate on two different vehicles. The Act would require documentary proof of citizenship to register for federal elections. Separately, the House adopted a budget resolution on July 22, by a vote of 216-214, instructing the committee that writes election law to report legislation adding up to $10 billion to the deficit over the next decade. The resolution does not say what the money buys. Supporters have described the plan so far as grants for states that adopt parts of the new requirements. That funding can move at a simple majority if it survives the Senate’s budget rules. The requirement itself, as ordinary legislation, still faces the 60-vote threshold, and at least one Senate Republican has already announced a no vote.

Congress could end up funding state implementation before it has defined what states are implementing. I have watched that sequence before.

If Congress is going to pay for this, it should decide first what it is paying for.

Start with what is not in dispute. Under 18 U.S.C. 611, noncitizen voting in federal elections has been a federal crime since 1996. It carries a fine, up to a year in prison, or both, and it can carry serious immigration consequences for a noncitizen. One detail worth knowing: that section covers the act of voting. Registration offenses are charged under other statutes.

States have found cases. Last week New Jersey’s governor said a software error in the state’s motor vehicle system registered roughly 6,600 people who had indicated they were not citizens, between June 2023 and June 2024, and that a preliminary review showed fewer than 400 of them voted. In January, Alabama’s secretary of state announced that his office had identified 186 noncitizens on the state’s voter file and removed all of them, and that voting history showed 25 of the 186 had cast ballots in Alabama elections. Michigan compared more than 7.9 million active driving records against more than 7.2 million active registrations and identified 15 people who appeared to be noncitizens and had voted in the 2024 general election. Thirteen were referred to the attorney general. One had died. One case was still open.

I have no basis to second-guess those findings and no interest in trying. They are small numbers against large rolls, but they are real.

Read the New Jersey one again, because it is not what people usually mean by noncitizen registration. These people were at a motor vehicle counter applying for a license or an ID. They were asked whether they wanted to register to vote. On that same keypad they indicated they were not citizens and not eligible. The software registered them anyway, and it did that for a year before anyone fixed it.

So, ask the question that matters. Why did the system accept a registration at all from someone who had just answered, on that screen, that they were not a U.S. citizen? An answer that disqualifies you should stop the transaction. It should not be a field that gets collected and then written past.

The vendor’s response is worth reading. It said its role was to transmit information through the motor vehicle system, and that the state’s Division of Elections was ultimately responsible for verifying eligibility and adjudicating what it received. That may be accurate. It is also how a determination goes missing. One party collects, another is supposed to decide, and the record gets created in between.

The failure did not start with a false answer. Nothing had been forged. The citizenship information was not missing, and it was not unclear. It was collected and then written past. This failure did not need documentary proof to expose it. The disqualifying answer was already in hand, and it should have stopped the transaction.

What interests me is the step before the finding.

Alabama’s earlier effort flagged 3,251 registrants who had been issued noncitizen identification numbers by the Department of Homeland Security (DHS). The secretary of state said at the time that some of those people may have naturalized since the number was issued, and his office set up a way for them to correct the record. The later effort, the one that produced 186, ran the file against a federal verification system, sent notice to each person identified, gave them a chance to submit documentation, and removed only those who did not respond.

New Jersey ran the same way. DHS had written to the state saying it might have tens of thousands of noncitizens on its rolls and reported flagging 35,152 names in the state’s public voter file. The state’s own review produced 6,600 registrations and fewer than 400 votes. Neither side has accounted for the difference.

Those are numbers of different kinds. One is a flag. The other is a determination. The work in between is the whole job, and it is the part nobody budgets for.

Get that part wrong in the other direction and you get Kansas. When Kansas required documentary proof, the district court found 31,089 applications suspended or canceled, and that more than 99 percent of the suspended applicants were citizens. The same record showed that over 19 years, at most 39 noncitizens reached the rolls. Kansas had an alternative process for people who could not produce a document. It was used five times. In Fish v. Schwab, the Tenth Circuit affirmed those findings and agreed the alternative added to the burden rather than relieving it.

Donna Bucci was one of them. She was born in Maryland, did not have a copy of her birth certificate, and testified that she could not readily afford the fee to order one.

One thing about Kansas that matters here. Kansas lost because a state was adding a requirement to a federal process. Arizona lost for the same reason in 2013. The Supreme Court held that the National Voter Registration Act of 1993 (NVRA) required states to “accept and use” the federal voter registration form, preempting Arizona’s attempt to require documentary proof of citizenship from applicants who used it. Congress amending the NVRA is a different question and a stronger one. I am not arguing that the proof of citizenship requirement is unconstitutional. I am arguing that the Kansas record shows what the requirement produced in practice.

Records do not always match. A national survey conducted in 2023 for the Brennan Center for Justice, the University of Maryland Center for Democracy and Civic Engagement, and VoteRiders estimated that 21.3 million voting-age citizens lacked ready access to documentary proof of citizenship. Someone may need to replace a birth certificate, reconcile a name change, or fix a database error. Sometimes the record itself is gone. Courthouse fires, floods, and storms have destroyed vital records, and in 1973 a fire at the National Personnel Records Center in St. Louis destroyed 16 to 18 million military personnel files, roughly 80 percent of the Army files for anyone discharged between 1912 and 1960. No duplicates or microfilm copies existed. This bill lists a military service record as one of its accepted documents.

Those are administrative problems. None of them makes a person ineligible. Resolve them once, with help, and they stay resolved. Leave them unresolved and they come back at every new registration: a move across a state line, a purge and a return to the rolls, a first registration at eighteen.

Asking a citizen to prove citizenship is a legitimate government function, and there is already a federal model for it. A passport application asks for that proof once, and the passport becomes the proof after that. State Department guidance says a previous passport works as proof of citizenship even when expired. This bill requires a valid one. An amendment to accept an expired passport was offered in the Rules Committee and never got a vote under a closed rule.

The analogy is not exact. A passport is adjudicated by one federal agency with specialists in it, and voter registration runs through thousands of state and county offices. The principle still holds. When the government has established something that does not change, it should keep a way to recognize that finding later.

A naturalized citizen has already proved citizenship. To the federal government, under oath, after a process that takes years. Asking that person to prove it again, with a document they may no longer possess or be able to obtain quickly, is asking the government to re-verify what the government already decided.

Citizenship does not change when you move. Residence, precinct, and registration do. A system built on the part that does not change would verify citizenship and identity once, keep a record of that determination, and give the citizen proof a state can accept later. States would still confirm residence and apply their own registration rules. A move from Florida to Georgia would mean a new state registration. It should not mean another search for the same birth certificate.

The bill’s first listed pathway is a REAL ID-compliant credential that also indicates citizenship. An ordinary one does not. REAL ID is not a separate card. It is a federal standard that a state driver’s license or identification card either meets or does not, and a marking, usually a star, shows that it does. Getting one means presenting documents that establish identity, lawful status, a Social Security number, and residence. Non-drivers can get the same card and the same marking. That could become one of the least costly pathways over time, because the counter already reviews documents and most people already renew on a schedule. A REAL ID establishes lawful presence, and a lawful permanent resident can hold one. Five states issue an enhanced license that does establish citizenship, because only citizens are eligible: Michigan, Minnesota, New York, Vermont, and Washington. Those carry a flag instead of a star. The flag already means what the bill is asking for. They built it for land border crossings, not for voting.

Ask why only five, because the answer matters. Those states did not build these cards for voting. They built them under a border-crossing program, the Western Hemisphere Travel Initiative, and each state had to negotiate a separate agreement with DHS. The card works at the border because it carries a radio-frequency identification chip that pulls the holder’s record from a federal database as the car reaches the booth. The chip drew heavy privacy objections, and it is one reason the model never spread.

So, the lesson is not that forty-five states were negligent. It is that the citizenship marking came bundled with surveillance infrastructure, and forty-five states declined the bundle. Unbundle it. A printed indication that a counter verified citizenship needs no chip that broadcasts, and no state would need a separate agreement with DHS to issue it. A printed marking by itself is still just a claim on a card. It needs signed machine-readable data alongside it, the way a passport chip carries signed data, so another state can confirm which office issued the marking and that it belongs to that card, without pulling up the holder’s records. Congress has already seen the drafting. A bill in the last Congress, H.R. 4597, would have amended the REAL ID Act to require an indication of citizenship on the card, with an explicit prohibition on including any other information about the holder’s immigration status. It did not pass. The language exists.

A federal commission proposed marking citizenship on the card in 2005. Voting rights groups objected then and have objected to similar proposals since. Twenty-one years on, most states still do not mark the card, and this bill lists the marked card first. If Congress is going to spend $10 billion, that is something worth buying. Verify once, at a counter that already verifies documents, mark the card, and let the card be the proof.

That marked license cannot be the only control. Cards get forged and altered, and a registrar holding one needs a way to confirm the determination behind it was real. So there has to be a record.

What matters is what the record holds. Enough to bind the determination to the right person and to show the credential is genuine: a status, a date, the office that made it, and a protected identifier. Nothing beyond that. No party affiliation. No vote history. No record of which elections a person voted in or how they voted. No address history. A registrar asks one question and gets one answer. Data minimization means keeping what the system needs to work and nothing past it. It does not mean pretending secure matching and correction can run on a bare yes.

That is the opposite of what just got built. The federal database states use to check citizenship is also called SAVE, for Systematic Alien Verification for Entitlements. It is a different thing from the Act. It was built to verify immigration status for benefit programs. The name says so. It is not a citizenship registry. The court that blocked the revamped version of that database from voter list checking found the government had combined and repurposed the personal information of millions of Americans, linked immigration records to Social Security data for the first time, and pulled in the records of people born here. A narrow record is a smaller target and a smaller temptation. It is not immune. A list of verified citizens has uses that have nothing to do with registering voters, which is why what may be done with it has to be written into the statute rather than left to whoever ends up running it. I am not proposing that Congress improvise a national citizenship registry inside this bill. I am saying Congress has to decide what determination is being kept, who can reach it, and how it gets corrected, before it funds anything.

A counter clerk should not be deciding hard citizenship cases. Some citizenship runs through a parent, or an adoption, or a statute that was in force on the day someone was born. Those need a trained adjudicator and a route to reach one. The clerk records a determination when a listed document establishes it plainly, and refers everything else.

I should say what is hard about what I am describing, because a proposal without its costs is not a proposal.

The first years would likely be slow and expensive. You would be enrolling a very large number of people at once, and the ones needing the most help are the ones whose records are hardest to find. Names change. Birth records get amended. Some records burned. Rural access, disability access and language access have to be solved rather than mentioned.

Then there is the architecture. Any record of who has been verified is a target, including the narrow one I just described. Minimizing what it holds limits the damage. It does not remove the risk, and whoever runs it will be under constant pressure to widen it, because a list of verified citizens is useful for a great many things that have nothing to do with registering voters. The only version of this I would trust is one where the limits are written into the statute instead of left to guidance. That is the same argument as everything else here. That is not a footnote. If the answer to it is bad, the answers to everything else stop mattering, and the June ruling on that database is a preview of what happens when a system gets built before that question is settled.

And the idea is not original to me. REAL ID is the model for a federal standard with a locally issued card, the marking proposal is twenty-one years old, and the bill itself points states at back-end checks. Others have written about this bill’s implementation problem. What I am adding is not the idea. It is the vantage point. I have worked the review side, opening the file after the money was already gone.

The bill does recognize the problem. It directs each state to establish a process for applicants whose documents contain a discrepancy, and an alternative process for those who cannot produce a listed document. It specifies what the official must record when the answer is yes: an attestation under penalty of perjury, an affidavit from the official, and a uniform affidavit form developed by the Election Assistance Commission (EAC) explaining why the person was registered. The bill does send notices. A mail applicant gets one telling them the requirement exists and how to meet it. Someone flagged during the roll screening gets one before removal, with a chance to produce documents. An applicant whose documents show a discrepancy can submit more. Those are real and they matter.

What the alternative process does not have is anything on the far side of a no. An official weighs the evidence and decides. If the answer is no, the statute does not require anyone to tell the applicant what fell short, give them time to fix it, or let a second person look.

An attestation under penalty of perjury is a self-certification. It is an enforcement hook, not a verification control. It tells you what to prosecute afterward. By itself it does not establish whether the person at the counter is eligible now.

I know what self-certification produces when a high-volume program treats the signature as the end of the determination instead of the beginning. The Paycheck Protection Program (PPP) ran that way. Applicants certified their eligibility and their numbers, the certifications were signed, and the money went out. I reviewed those files afterward.

That is not an argument against the attestation option. For a citizen whose records burned or lost in a natural disaster, it is the only thing standing between them and no registration at all. It is an argument that a self-certification is where adjudication starts, not where it ends. Somebody has to look. The bill leaves who, and how, and to what standard, to whoever is standing at the counter.

There is one cure period in this bill with a number attached. A voter who arrives without photo identification can cast a provisional ballot and has three days to produce the ID or sign an affidavit stating a religious objection to being photographed. The statute says three days. It does not say business days. Unqualified days in a federal statute normally means calendar days. Unless implementing guidance supplies a different computation, that would mean you vote on a Tuesday and you have until Friday.

Now set that beside another number in the same bill. Federal agencies get twenty-four hours to answer a state’s request for verification information. So, the bill sets a deadline for what the government owes a state, and a deadline for what a citizen owes the government, and no deadline for what any agency owes a citizen who is trying to obtain a document. If you are waiting on a county vital records office, the wait is your problem, and your three days are running.

And that three-day window is for photo identification. On the citizenship side there is no stated window at all.

The bill does create exposure for the official who says yes. It adds registering an applicant without documentary proof to the private right of action in the NVRA, which carries no intent requirement, and to the criminal penalties, which sit inside a provision reaching a person who knowingly and willfully defrauds residents of a fair and impartially conducted election. So, an honest mistake is probably not a crime. Whether it becomes a lawsuit depends on standing, on the Act’s notice requirements, on who gets named, and on immunity. What this section does not do is create anything comparable when an eligible applicant is wrongly turned away. Other law may reach that. This section does not. The incentives are not symmetrical. Kansas had a process too.

Now read what the bill says about getting all this running.

Section 2 takes effect on the date of enactment. It applies to applications submitted on or after that date, so people already registered are not asked to re-register. That limit is real and supporters are right to point it out.

The EAC gets ten days to write implementation guidance and transmit it to every state’s chief election official.

Within thirty days, every state has to establish a program that submits its complete list of registered voters to DHS through SAVE, identifies people who are not citizens, and takes the steps necessary to remove them, after notice and a chance to provide documentary proof.

Read those two together. If you are already registered, you do not fill out a new form. You do get run through the database, and if the comparison flags you as a possible noncitizen, you produce documents or you come off the roll. The grandfather clause protects you from the paperwork. It does not protect you from the screen.

That is the same system a federal judge blocked from voter list checking last month, on a record that included one state flagging actual citizens as possible noncitizens. The court wrote that the government had repurposed citizenship data it knew to be unreliable. The government has paused some features to comply and says it will keep defending the system.

Federal agencies get twenty-four hours to answer a state’s verification request, including batched requests.

And Section 2(m) makes the Paperwork Reduction Act (PRA) inapplicable to the development or modification of voter registration materials, including the forms.

That last one is the provision I would have flagged in a review. The Paperwork Reduction Act is what makes an agency estimate the burden a form puts on the public, publish it for comment, and get Office of Management and Budget approval before running a covered collection. It is the floor. The bill removes it for the development or modification of the registration materials this program runs on.

Legislation establishes the policy. Implementation determines whether it works.

Congress has more options here than it usually uses. It can state a policy and leave the rest to agencies. It can write implementation requirements into the statute. It can do both, and it often does.

When it does the second, it looks ordinary: a deadline that reflects how long the work takes, a list of subjects the guidance has to cover, a requirement to consult the people who will run the program, a funding authorization, reporting back to Congress, and a Government Accountability Office (GAO) or Inspector General review after the fact. None of that is unusual. It is in a lot of federal law.

What it does is set a floor. The agency still writes the manual, the forms, and the procedures. It just cannot decide on its own whether a correction process is worth having.

Instead of directing states to establish a process, a statute can say what the process has to address.

Before any of it starts: enough time to build the thing before the requirement takes effect. Ten days and thirty days are not schedules, they are announcements. A program like this needs standards, a pilot, state grants, testing, and a compliance date set after the pathway exists instead of before it, which is election cycles rather than weeks. A burden estimate and public comment on the forms, which is what the PRA exists to force and what this bill switches off. A rule that a database match is a reason to look and not a finding, and that nobody comes off a roll on a match alone. And a named office that owns the determination, so it cannot fall between a vendor that transmits and an agency that assumed somebody else was deciding.

Then the correction path. Notice of what could not be verified. A period to respond, measured against how long those records actually take to obtain. Help retrieving them, at no cost to the applicant. Deadlines that run against the agencies too, not only against the citizen. Review by someone other than the person who made the first call. Training for the people making determinations. A record that survives an audit, with statutory limits on what it may contain. Protection for an official who makes a good-faith determination in either direction. A rule that nobody may draw any conclusion from a card that lacks the marking. And a determination that stays valid for later registrations and interstate moves unless the government develops specific, reliable evidence that it was wrong.

The Act is specific about what states must require. It leaves critical parts of the operation to them.

I watched what happens without a floor. During PPP, applications and money were flowing while the guidance and the review procedures were still being written. GAO documented the limited initial safeguards and the consequences for program integrity. The program worked in the sense that money reached businesses. Lenders confirmed the certifications and the payroll documents, and they were entitled to rely on what the borrower represented. Whether the borrower was actually eligible was a question the file did not test until forgiveness.

Voter registration is a different function. The sequencing concern is the same.

If this becomes law, restoring confidence will not turn on the requirement. It will turn on whether states get enough time and enough money, whether the people making determinations are trained, whether the systems work, and whether an eligible citizen whose records do not match has somewhere to go. None of that happens by itself. Congress can require it, or Congress can leave it to whoever is standing at the counter.

Voting is a right of citizenship, whether that citizenship came at birth or at a naturalization ceremony. Elections have to be secure and they have to be fair. I am not arguing with any of that. I am arguing that the operation has to be built to match.

So look at the size of the job. More than 211 million people were active registered voters for the 2024 general election, by the states’ own reporting. Every one of those records goes to DHS for screening within thirty days of enactment. About 870,000 people naturalize in a typical year and become newly eligible. More than 21 million registration records came off state lists in the single two-year cycle between the 2022 and 2024 elections, and many of those people register again somewhere. And motor vehicle offices and automatic registration are the two most-used registration methods in the country, which means the channel that failed in New Jersey is the busiest one there is.

That is the work. It gets done by county clerks and motor vehicle staff, on deadlines Congress sets, with money Congress has not defined.

If we are going to do this, we should want it done right, and that is not a soft position. A transition that works costs less than one that fails. A program that has to be rebuilt gets paid for twice, once to stand it up and once to fix it. In my experience, the repair is never the cheaper half. That is not a prediction. I reviewed the files.

Congress is about to decide what to spend on proof of citizenship. It should decide at the same time whether proving it once ever counts for anything.

Traci Harig is a former GS-13 loan specialist whose work at SBA included eligibility and forgiveness reviews, affiliation analysis, appeals, and matters referred to agency counsel, DOJ and OIG. She writes about program design, eligibility verification, and federal program integrity.

This is an independent analysis based on publicly available information and professional experience. The views expressed are my own. I have no clients and no financial or advocacy interest in the matters discussed.

Sources & References

1. U.S. House Committee on Rules. H.R. 8595, 119th Congress.
2. U.S. House Committee on Appropriations. Press release on House passage of H.R. 8595.
3. U.S. House Committee on Rules. H.R. 8800, 119th Congress.
4. U.S. House of Representatives. Rules Committee Print, S. 1383, SAVE America Act, bill text.
5. U.S. House of Representatives. Committee document posting, budget resolution.
6. Committee for a Responsible Federal Budget. Analysis of the House FY 2027 budget resolution.
7. The Hill. Coverage of the House reconciliation budget resolution.
8. The Wall Street Journal. Opinion on voting against the SAVE America Act.
9. 18 U.S.C. 611, Voting by aliens. Office of the Law Revision Counsel.
10. NBC News. Report on non-citizens registered to vote in New Jersey through a motor vehicle software error.
11. Alabama Secretary of State. Press release directing removal of 186 noncitizens from the voter file.
12. Michigan Department of State. Review of noncitizen voting in the 2024 general election.
13. New Jersey Monitor. Report on the software error and noncitizen registrations.
14. Alabama Secretary of State. Press release implementing a process to remove noncitizens registered to vote.
15. Fish v. Schwab, 957 F.3d 1105 (10th Cir. 2020).
16. Arizona v. Inter Tribal Council of Arizona, 570 U.S. 1 (2013).
17. Brennan Center for Justice. Analysis of the 2023 survey on access to documentary proof of citizenship.
18. National Archives and Records Administration. The 1973 fire at the National Personnel Records Center.
19. U.S. Department of State. Passport renewal guidance.
20. Transportation Security Administration. REAL ID frequently asked questions.
21. U.S. Department of Homeland Security. Enhanced driver’s licenses, what are they?
22. H.R. 4597, 118th Congress, bill text.
23. NPR. Report on the court ruling on use of the SAVE database for voter list checking.
24. U.S. Government Accountability Office. GAO-21-577.
25. U.S. Election Assistance Commission. Release of the 2024 Election Administration and Voting Survey.
26. Office of Homeland Security Statistics. Naturalizations annual flow report.

← Back to Policy Analyses | Home