Restricted analysis, made public daily.
Declassified under standing order Edition No. 125 Thursday, November 5, 2026

Over 20,000 People Remained
on the List, a Declaration Says

In an order denying a temporary restraining order, a federal judge says the government’s own description of how it compiled the October List indicates a process that involved the “matching of voters to their immigration status.”

Judge David O. Carter’s order of Oct. 28, 2026 denies the application of the United States, the plaintiff, for a temporary restraining order in United States of America v. State of California et al., in the U.S. District Court for the Central District of California. Eric McLoughlin of Homeland Security Investigations (HSI) says in a declaration that on Oct. 13, 2026 he helped facilitate delivery of “that list of just over 20,000 individuals who HSI believes are (1) registered to vote in California, (2) were sent a vote-by-mail ballot to participate in the federal elections, and (3) are noncitizen aliens.”

The order calls it “the October List” and says: “Given the vagueness of Plaintiff’s descriptions of its procedures and the unreliability of its similar claims in other states, the Court cannot conclude that Plaintiff’s process is credible.” It adds that this “calls into question” the plaintiff’s allegation that the individuals on the October List are “highly likely to be noncitizen[s].”

What the declarations say

Nicolas Bartell is the Associate Director of the Fraud Detection and National Security Directorate (FDNS) at U.S. Citizenship and Immigration Services (USCIS). In paragraph 16 of his sworn declaration, filed Oct. 21, 2026, he states that “USCIS then scores and ranks the potential identity match.” He states the matches are scored and ranked because the resulting data may be incomplete, outdated, or mismatched and the potential matches need to be prioritized for manual review. In paragraph 17 he states that USCIS screens all potential matches against Department of State and Social Security Administration data to remove individuals who can be identified as United States citizens or as deceased.

USCIS then consolidates all match evidence and assigns a confidence level for each match, he states in paragraph 18; in paragraph 19, he states that FDNS investigators manually review the matches to confirm the identity match. A second-level review of the match and citizenship status determination, he states in the same paragraph, occurs “as an additional safeguard.” In paragraph 22 he says FDNS took the California voter list and put it through “the process identified above.”

McLoughlin is the Assistant Director for the Cyber and Operational Technology section at HSI. In paragraph 17 of his declaration, filed Oct. 21, 2026, he states that HSI’s enrichment process focuses on improving the completeness and consistency of identity and address information and, where lawfully permitted, identifying additional unique identifiers that may assist in “reliable matching of the voters to their immigration status.”

In paragraph 15 of that declaration he says, “As I understand, FDNS developed a model that uses information from the voter roll and combines it with information from a reliable third-party data analytics and investigative platform to learn more about the individuals.” He states in paragraph 23 that HSI’s further refinement “resulted in over 20,000 individuals remaining on the list.” As of Nov. 4, 2026, neither of the two government declarations we read names the platform; both call it “a reliable third-party data analytics and investigative platform.”

Aman Bhullar says he is the Assistant Registrar-Recorder/County Clerk of the Information Technology Bureau of the Los Angeles County Registrar-Recorder/County Clerk, responsible for election management and tally systems. In a declaration filed in support of the Registrar’s opposition, he says that, at counsel’s request, his team conducted a preliminary analysis of a sample of two hundred records from the October List. He says 42.9% of the sample had canceled voter registration records in the county’s system, 42.4% had active records, and 4.6% are considered inactive records. The order recites the same three figures from that 200-record sample. He adds that 7,035 of the more than 20,000 records on the October List have addresses in Los Angeles County.

What the order says

The order’s section heading says the United States’ requested relief “would likely violate federal election law,” and its sub-heading says the remedy “would likely violate the National Voter Registration Act of 1993.” Both headings concern the relief the United States asked for. Under that sub-heading the order quotes the Act as requiring states to complete, not later than 90 days before a federal primary or general election, “any program the purpose of which is to systematically remove the names of ineligible voters from the official lists of eligible voters.” The order quotes the Ninth Circuit’s Mi Familia Vota decision, which the Supreme Court has agreed to review, on “a mass computerized data-matching process.” The order says: “Plaintiff’s own description of how it compiled the October List likewise indicates a process that involved the ‘matching of voters to their immigration status’ by relying on information from various databases, including USCIS, Department of State, and Social Security Administration data.”

The plaintiff “waited until the night before the hearing to reveal that the third-party platform it relied on was Lexis Nexis,” the order says. The order cites a second declaration from Nicolas Bartell, filed as a supplement to the plaintiff’s application, which we did not have. Of that platform, the order states: “It remains unknown how Lexis Nexis performed its analysis.” The plaintiff “likewise did not explain what its ‘enrichment process’ entailed,” the order adds, or how it occurred. The plaintiff provided “no error rates” or “validation against a reliable benchmark,” the order says, citing a declaration filed with the LULAC and NAACP opposition. The order says: “Omissions of this critical information prevent an evaluation of Plaintiff’s procedure, including how Plaintiff reduced its initial claim of nearly 200,000 alleged noncitizens to 20,000 individuals.”

What has followed in court

The Ninth Circuit’s order says the United States immediately appealed and filed an emergency motion for an injunction pending appeal on Oct. 29, 2026. The Ninth Circuit’s published per curiam order of Oct. 30, 2026 denies the motion, saying the United States “has failed to meet its burden in showing that an injunction pending appeal is warranted.” The same order says the potential violation of the National Voter Registration Act “provides an independent reason to conclude that the United States has not shown a clear likelihood of success.” Judge Patrick J. Bumatay dissented. On Monday, Nov. 2, 2026, the Supreme Court denied the application for an injunction pending appeal in No. 26A624. As of Nov. 4, 2026, the order text we read states no reasons and records no noted dissent.

The district judge’s order of Oct. 29 appoints Daniel B. Garrie as “a technical Special Master to assist the Court in developing the factual record.” It also makes witness lists, including expert witnesses, due by Friday, Nov. 6, 2026, “regarding the reliability of the October List and the methodology Plaintiff used to produce it.” It sets a hearing for 10:00 a.m. on Friday, Nov. 13, 2026, on the State and County motions to dismiss and the plaintiff’s motion for a preliminary injunction.

Two levels of review were sworn to. “No error rates,” the order says the plaintiff provided.

The Agentics read

Human review is the safeguard readers expect, and the Oct. 28 order recites one for each match, from the government’s own description; the same order says the plaintiff provided “no error rates.” A review is a step; an error rate is a count, and the order says the plaintiff provided none. When a system judges you, ask for the count, not the description.

What each document says.

What the order says. The plaintiff “did not explain what its ‘enrichment process’ entailed,” the order says, adding that it remains unknown how the third-party platform performed its analysis. The plaintiff provided “no error rates,” the order says, citing a declaration filed with the LULAC and NAACP opposition.

What the declarations say. A USCIS official’s declaration says USCIS “scores and ranks the potential identity match” and that investigators manually review the matches.

What comes next. Witness lists on the reliability of the October List and the methodology the plaintiff used to produce it are due Nov. 6, 2026, the Oct. 29 order says. A hearing is set for Nov. 13, 2026.

Where the record stops

As of Nov. 4, 2026, none of the court documents and declarations this edition read names, or contains a statement from, a person identified as being on the October List; the LULAC and NAACP opposition says that some of the roughly 20,000 “may be” members of those groups, naming no one. We did not read the complaint, the application for the order, the second Bartell declaration the order cites, the declaration the order cites for “no error rates” or the Ninth Circuit dissent.

The denials here are early-stage rulings, not final judgments, and the case is live. That USCIS scores and ranks potential identity matches rests on the government’s own sworn declaration. The order says it remains unknown how the third-party platform performed its analysis. Nothing here predicts the outcome.

Through Nov. 4, 2026, we contacted none of the parties. Statements about the record here are attributed to the document that makes them. This page reports the filings and is not a finding about the court, the parties, the declarants, the platform or any person on the list.

Sources

The record (public court record). United States District Court for the Central District of California, United States of America v. State of California et al., No. 2:26-cv-12025-DOC-ADS: order denying the application for a temporary restraining order, Dkt. 149, Oct. 28, 2026; order appointing a Special Master, setting a hearing and setting a briefing schedule, Dkt. 159, Oct. 29, 2026 (unlinked); declaration of Nicolas Bartell, Dkt. 62-2, filed Oct. 21, 2026 (unlinked); declaration of Eric McLoughlin, Dkt. 62-3, filed Oct. 21, 2026 (unlinked); declaration of Aman Bhullar, Dkt. 32-3, filed Oct. 20, 2026, in support of the Registrar’s opposition (unlinked); the LULAC and NAACP opposition, Dkt. 37-1, filed Oct. 20, 2026 (unlinked). United States Court of Appeals for the Ninth Circuit, per curiam order, No. 26-7231, Oct. 30, 2026 (unlinked). Supreme Court of the United States, order, No. 26A624, Nov. 2, 2026 (unlinked).

Our read: a ranking decides whether you’re seen.
Has anyone told you its error rate?

A step can be sworn to; an error rate has to be counted. The Oct. 29 order makes witness lists on the methodology the plaintiff used to produce the October List due Nov. 6, 2026. Name one ranking your work depends on, and write down its error rate. If that line stays blank, bring that question to SIA™, the Intelligence Officer.

Every edition, in order, from No. 001 · A new edition releases daily, 05:30 CT.

Open the Record
‹ Edition No. 124 Next Dispatch — releases 05:30 CT ›