The credibility of this referendum campaign rests on the credibility of its evidence. We hold ourselves to a higher standard than partisan advocacy requires, because we believe the case for our position is strong enough to be made plainly. We expect to be examined. We have tried to make examination easy.
Trexit.org asks Americans to make the November 3, 2026 elections a referendum on the administration's autocratic governance: to vote, to bring two eligible voters with them, and to vote against any elected official, of any party, who they believe has enabled the administration's corruption and abuse of power. That ask rests on a factual record with three parts: false statements made publicly by the subject and contradicted or shown to be materially misleading by verified evidence, documented incidents of corruption and self-enrichment, and documented incidents of abuse of official power. The credibility of the pledge depends on the credibility of the record. This page explains, in detail, how the record is built.
We hold ourselves to a higher standard than partisan advocacy requires, because we believe the factual case for our position is strong enough to be made plainly, without exaggeration. We expect our methodology to be examined by journalists, by political opponents, and by our own signers. We have tried to make examination easy.
Trexit is a project of Protect American Democracy, a 501(c)(4) social welfare organization, funded by voluntary contributions. It is not controlled by and does not coordinate with any campaign, political party, candidate, or campaign committee, and contributors have no editorial role. More about the project, how it is funded, and the disclaimer that governs it, is on the About page.
The fact-checks and incident records published here are compiled from publicly available sources, each of which is cited. The characterizations and conclusions are those of Protect American Democracy, offered as political expression. Its staff and directors write in that capacity only, not for any employer. We hold the work to a high standard and invite scrutiny: read the sources, weigh the evidence, and judge it for yourself.
The Record documents three kinds of conduct in the president's second term: false claims he has made, corruption, and abuse of power. Each entry is built from verified evidence, scored by the Demos verification engine under one fixed configuration, and tiered under the published rules below. An editor confirms every tier.
Every change to the Record, and the reason for it, is listed in the changelog.
The Record covers the president's second term, from January 20, 2025. Every entry meets four tests.
Fact-check articles are cited on each entry as sources readers can follow, but they are not used as evidence when an entry is scored. The engine scores the underlying record itself.
The Demos verification engine reads an entry's claim and its verified evidence, and produces two scores from 0 to 100.
Each entry is scored six times and the median of each score is used, so no single run decides a tier. An entry whose scores land within 2 points of a threshold is scored six more times. If it is still that close, the editor decides, and the decision is recorded.
Every entry was scored under one fixed configuration: the same model, the same instructions and the same scoring geometry. Each entry records that configuration and the date it was scored, and publishes its V and C.
Every published entry is Tier A or Tier B. No entry is published unless its evidence reaches a confidence (C) of 50.
False claims. The editor sets each false claim's tier; the engine screens it.
We chose editorial tiers for false claims because of the test described below. On a blind sample of false claims, the engine's tiers did not agree with an independent review more often than our editors' tiers did, so the engine does not set them. It still decides whether a statement's evidence is strong enough to publish, and where its reading differs from the editor's, the entry is reviewed again.
Corruption and abuse of power. The engine sets the tier, and the editor confirms it.
What must be established depends on the category. For corruption, it is value flowing to the president or his family; for a pardon, or an official act that benefits a donor, a documented return to the president, such as his own stated reason or a donation. A pardon meets the same test wherever the Record files it. For abuse of power, it is the unlawful or retaliatory use of governmental power, including a coercive threat to use it. An act alone is not enough: an executive order proves the order exists, not that it was corrupt.
An editor reviews every entry before it is published: most closely, every entry the engine could not place, every close call, every statement whose checkability is in question, and every entry where the engine's reading differs from the editor's. For each, the editor confirms the tier, sets a different one, or withdraws the entry.
The scoring rule, and the test it has to pass, were fixed in writing before the test was run.
We drew 30 entries at random, 10 per category. Blind reviewers, who saw neither the engine's result nor any existing tier, tiered each entry from its evidence, and the editor decided the correct tier for each before seeing the engine's. In each category, the engine sets tiers only if its tiers matched the editor's rulings at least as often as editorial tiers did.
| Category | Engine tiers matched | Editorial tiers matched | Outcome |
|---|---|---|---|
| False claims | 5 of 10 | 6 of 10 | The editor sets tiers; the engine screens |
| Corruption | 3 of 10 | 3 of 10 | The engine sets tiers; the editor confirms |
| Abuse of power | 5 of 10 | 5 of 10 | The engine sets tiers; the editor confirms |
Where the engine could not settle an entry and referred it to the editor, we counted that as a miss. Ten entries per category is a small sample, and the margins are narrow. That is why every tier, whichever method set it, is confirmed by an editor.
A false claim is dated to the day the president made the statement. Conduct is dated to its first official act within the term. Where that act's date cannot be established, the entry is dated to the first report of it and marked accordingly. Later rulings, reports and disclosures never change an entry's date. Dates are checked to within about three months, which is the precision the Record's month-by-month view needs.
The changelog lists every change to the Record with its reason: every tier that moved, every correction to an entry's text or date, every entry merged with a duplicate, and every withdrawal.
A tier describes the strength of the evidence, not the seriousness of the conduct. A Tier B corruption entry may be as grave as a Tier A one; it means no official record has yet established it. A tier is not a legal finding. No entry states that anyone committed a crime unless a court has found so, and outcomes on both sides of a case, such as rulings, stays and appeals, are stated.
The engine is a tool, and its scores depend on the evidence it is given. That is why every source is verified before scoring, and why an editor confirms every tier.
Rescoring. Entries are rescored only on a published schedule, with a changelog, never automatically when the engine or its model changes. The current configuration stays fixed until the next scheduled review.
The corruption and abuse-of-power records are governed by rules that we apply before we know what they will produce. An incident is eligible only if all of the following are true.
We exclude, deliberately and by rule:
Two further rules govern how an incident is written. Where a district court ruled one way and an appellate court stayed or reversed it, both are stated; we never collapse a two-sided outcome into the half we prefer. And characterizations belong to the people who made them. Where a record calls something an abuse of prosecuting power, a false pretense, or a thinly veiled threat, those are the words of a named judge, quoted as such. They are not ours.
A record of this size invites a fair question: how do you know you have not simply collected the incidents that suit you, and passed over the ones that do not? We take the question seriously enough to have tried to answer it, and honest enough to tell you where the answer falls short.
Over eight sessions we swept a roster of fifty-five sources spanning eight independent classes: litigation trackers, legal scholars across the ideological spectrum, inspectors general, the Government Accountability Office, congressional oversight products, campaign finance filings, executive-branch financial disclosures, and press-freedom and human-rights monitors. Fifty-one were harvested and classified. Three could not be reached. One was left unfinished.
The single most useful result concerns litigation. The Just Security Litigation Tracker maintains a database of federal court challenges to this administration's actions. We harvested its Blocked family: two hundred and sixty-eight federal actions that a court has blocked. Two hundred and sixty-four of them were already in this record. Four were not, and were added. That is roughly ninety-eight and a half percent coverage of the court-blocked universe, measured against a source we did not build and do not control.
We then ran three further litigation trackers against the same universe, independently: the Associated Press lawsuit tracker, the Civil Rights Litigation Clearinghouse, and the Court Accountability judicial-threat tracker. Each returned nothing this record did not already contain. Three trackers, three sets of editors, three sets of incentives, converging on the same set of incidents, is a stronger claim about completeness than any number of records could be on its own.
The exclusions matter as much as the inclusions. Six candidate incidents were dropped because controlling authority went the government's way. Sixteen of the seventeen suits we identified in one lane, cases the administration brought against states, were excluded as routine preemption litigation rather than abuse. Two executive orders that dominate public argument, on birthright citizenship and on transgender military service, are absent from this record because they are constitutional disputes still being litigated, not documented abuses. A record that included everything would tell you nothing.
Now the gaps, stated plainly, because a completeness argument that hides them is not one.
What the sweep establishes is not completeness. No sweep could. It establishes convergence: that eight classes of source, with different authors, different formats, different politics, and different blind spots, point at the same set of incidents. The marquee court defeats, the documented prosecutions, the grant conditions struck down, the disclosed income streams, the deaths during enforcement operations. Each recurs across sources that share nothing but the underlying facts.
One thing this record does not do is compare. Nobody has swept the record of any previous administration to this standard, including by us. The number of incidents documented here therefore supports no claim, and is offered as no claim, about whether this administration has done more of this than another. It is a count of what we found, in one place, against a stated standard, with the gaps named. What it means is for you to decide.
We draw evidence from sources that meet standard journalistic criteria for authority and independence. These include:
We do not rely on anonymous sources for primary factual claims, on partisan advocacy organizations as authoritative voices, or on social media reports without independent corroboration.
We hold ourselves to the following limits, and we ask our readers to hold us to them:
If you believe a fact-check on Trexit is inaccurate, we want to hear from you. Send your concern to contact@trexit.org with a subject line beginning "Correction:", and include the URL of the fact-check, the specific factual point you believe is wrong, and any evidence you can provide. Our editorial team reviews every correction request. We will respond, in writing, within seven business days. If your evidence supports a correction, we will make one. If it does not, we will explain why.
This page concerns our fact-checking methodology. Our privacy and data practices for pledge signers are described in detail on our Privacy Policy page. In brief: we do not display individual pledge-signer names publicly without explicit opt-in, we do not sell pledge-signer data under any circumstances, and we honor data deletion requests promptly. The pledge count we display reflects confirmed, deduplicated pledges (those who completed email confirmation) only, and the counting methodology is published. We will never display a count we cannot independently verify.
Trexit is funded by voluntary contributions from those who want this record maintained, made to Protect American Democracy. It carries no advertising and runs no sponsorships. Trexit accepts no contributions from foreign nationals, from any candidate committee, or made in the name of another person. Every contribution from an organization is published, including from political committees. No contributor of any kind has an editorial role. Contributions to Protect American Democracy are not tax deductible as charitable contributions.
Funding and editorial decisions are separate. Entries are selected, verified, tier-classified, and published on the evidence alone. No contributor, and no consideration of what any contributor might prefer, enters the scoring or the editorial review. Contributions are never solicited on the basis of a specific entry's content before that entry is published.
Some readers will ask why we publish fact-checks on a campaign site at all, rather than in a standalone research outlet. The answer is straightforward: the factual record we are building is the basis for the pledge's call to action. We believe that documented patterns of factual misstatement by a holder of high office are legitimately relevant to the public's evaluation of whether that officeholder should remain in office. We do not ask you to take our word for that. We ask you to read the fact-checks, evaluate the evidence, and decide for yourself.
The pledge itself calls for accountability through the most fundamental constitutional mechanism there is: the vote. Signers commit to vote in the November 3, 2026 election, to bring two eligible voters with them, and to apply a published, conduct-based standard (documented falsehood, votes against lawful oversight, the excusing of self-dealing) to every incumbent and candidate on their ballot, regardless of party. We do not advocate for any extra-constitutional action. We name no candidate and endorse no candidate. Our position is that the public is entitled to know what has been said and what has not been true, and to act on that knowledge where the Constitution says verdicts are rendered: at the ballot box.
Last updated: October 2, 2026 · Changelog
Contact: contact@trexit.org (for a correction, start the subject line with "Correction:")
Our methodology exists to be examined. If it earns your trust, the next step is to read the record it produces.