About Hypocrisy Now

Politicians change their positions. That is not automatically dishonest, and this site is not built on the idea that it is. What this site documents is the narrower and more useful thing: cases where someone argued a principle when it served them and abandoned that same principle when it did not.

The standard here is receipts, not rants. Every finding puts the earlier statement next to the later one, with dates and sources for both, so you can read them together and judge for yourself. If a claim cannot be sourced, it does not get published.

Findings start from public reporting and primary sources. Each one pairs two documented statements or actions from the same figure, and each side of the pair carries its own quote, date, and link back to where it came from. A finding with a missing receipt is not a finding.

Every published item is scored for severity on a scale of 0 to 10. The score reflects how directly the two positions conflict and how much the reversal mattered, not how much anyone dislikes the person. Scores are visible on every card so you can disagree with one.

Two rules constrain what gets covered. Public figures only: elected officials, candidates, appointees, party organizations, and people who have entered public debate on their own initiative. Private citizens are never included, including private citizens who happen to be related to public ones. The same yardstick for everyone: the detection process does not know or care which party a figure belongs to, and the same criteria are applied across the board. Where the resulting counts come out uneven, that is a finding in itself rather than something to correct for.

The research and writing on this site is done by AI, and the publishing system can put a finding on the site without a person seeing it first. An automated pipeline reads public sources, proposes candidate findings, drafts the comparisons, and scores them. A second automated stage then reviews each draft and sends it down one of three paths. When it recommends publication and both its own confidence and its quality score clear fixed thresholds, the finding is approved and published with no human involved. When it recommends publication but either score falls short, the finding stays in a review queue for a person to decide. When it recommends rejection, the finding is rejected outright whatever those scores say, so it is never published and never waits for anyone.

A human review tool exists and is used: any finding can be approved, rejected, pulled from the site, or published by hand through it. What is not true, and what this page previously said, is that nothing reaches the site without human review. The automatic path above does not require it. Saying otherwise would claim an editorial check the publishing path does not perform.

How the published archive divides between those three paths is not something this page can tell you, because no count has been published. Read the disclosure above as a description of what the system is allowed to do, not as a measurement of how often it did it.

We say this plainly because a site that asks you to check its receipts has no business being vague about its own. Knowing how the work is made is part of being able to evaluate it.

It also means the failure modes are worth naming. Automated research can misread context, miss the reason a position legitimately changed, or pair two statements that are less similar than they look. The sources are published on every finding for exactly this reason. Read them. If something here is wrong, the receipts are what will show it.

Corrections matter more than being first. If a receipt turns out to be wrong, misdated, or stripped of context that changes its meaning, the finding is corrected or removed from the site. Removal is a real state in the publishing system, not a figure of speech: a published finding can be pulled, and a pulled finding stops being served.

There is currently no published address or form for reporting an error. That is a gap, and naming it is more useful than pretending otherwise: corrections today are found during internal review rather than submitted by readers. There is no response-time commitment to make, because there is no intake channel to make one about. When a reporting route exists, it will be published here and this paragraph will say so instead.

There is also no correction log yet. Until one exists, do not read the absence of corrections on this site as evidence that none were made.

The hypocrisy feed carries every published finding. Politicians groups those findings by figure, and Who's Funding What follows the money behind the message.