1Drop a link
Paste a link, or paste the text itself. The crawler reads the page and eats the claims it can’t source, in front of you. Today that is a read of how each claim is written; taking them to the sources comes next.
Send a crawler →an AI fact-checker · being built · the crawler is live today
UNCITED is an AI fact-checker. It reads a page, goes to the source behind every claim, finds sources where none is named, and shows its work. The crawler is how you watch it: it eats what doesn’t hold up. We’re building it now. Today’s crawler is the first piece, and it runs on plain rules.
the page is invented. the crawler is real. the sources and verdicts are written by hand: a preview of the checker.
How it checks being built · shown below as an illustration
UNCITED is meant to do what a careful human fact-checker does, on every claim, without getting tired. This is the process, and then the process running on one normal paragraph.
The AI reads the text and picks out each claim.
If a claim names its source, the AI goes there, opens it, and reads what it really says.
If a claim names none, the AI searches and finds sources itself.
It compares what the sources say with the claim: backs it, backs part of it, contradicts it, or says nothing.
The crawler eats what doesn’t hold up and leaves the rest. The sources it used stay attached.
Illustration. The town, the newspaper, the council and every source and number here are made up, written by hand to show the process. The checker is not built yet. What exists today is the crawler, which reads how a claim is written and cannot open or search anything.
Illustration. The pages, the token and the sources are made up, written by hand to show the shape of the checker’s answer. The “amazing facts” are real, well-known facts, and the sources are described by kind (a survey agency, a weather service), not quoted. The checker is not built yet.
Two ways in
Paste a link, or paste the text itself. The crawler reads the page and eats the claims it can’t source, in front of you. Today that is a read of how each claim is written; taking them to the sources comes next.
Send a crawler →Open any page. If a crawler has already been there, it’s clean the moment you arrive. If not, the crawler cleans it for you, live, the way it would a link you dropped. Then it remembers, so the next person gets it clean.
Illustration. The page is made up, and the eaten sentences are today’s crawler (rules). The extension, the memory and the source check are not built yet; step 1 is what the crawler does today.
The crawler, explained
Feeds on claims nobody can source. If it isn’t sourced, it gets eaten.
Fig. 1. The adult, from above. Eight legs of three bones each. The body goes nowhere unless its legs haul it.
Fig. 2. The specimen on a page of hype (made up). The links are sources; the rest is prey.
Fig. 3. Four sentences from the page in Fig. 2, and what today’s crawler saw in each.
Every sentence of a page, in order, as a person would.
Today, by how a claim is written: a source, or loudness, absolutes, a number nobody backs. With the checker, by what several sources say.
What cannot stand on a source is shattered where it stood, with a margin note saying why.
Field note: this specimen is the picture of the checker being built, and today it runs on plain rules, not a language model. The same page gives the same result every time, and it says “unsourced”, never “false”.
The plan
The coin’s creator fees pay for what checking costs: the AI’s reading, the searches, the servers, the minutes it takes to do a page properly.
Crawlers work through the web all day, so more pages are already clean when you get there.
The extension shows where the crawlers are, and which pages they’ve already cleaned.
None of this exists yet. Today there is one crawler that reads how claims are written, and a link box. The checker, the coin, the fees, the swarm and the extension are the plan.
cleaned pagecrawler passed throughnot visited
Status
Send a link or paste text. It reads how each claim is written and eats the ones with no source.
Goes to the source behind each claim, finds sources where none is named, shows what they say, and the crawler eats what they don’t back.
Runs on the page you’re reading. Cleans it live if nobody has.
Cleaned pages remembered, so the next reader gets them clean.
Crawlers that work the web on their own, paid for by the fees.
Every payment from the fees, public, before anyone is asked to trust it.
Corrections
Yes. That is what UNCITED is: an AI fact-checker, and it is being built now. The AI reads a page, goes to the source behind each claim (the one the text names, or ones it finds itself) and says what the sources back and what they don’t. What exists today is the first piece, the crawler. It runs on plain rules and cannot open or search anything yet, so until the checker is live it says “unsourced”, never “false”.
The checker is: an AI reading the sources and weighing them against each claim. Today’s crawler is not. It is plain, written rules, the same page gives the same result every time, and the margin note says why each sentence was eaten.
Both, and it says which. When a claim names its source, the AI opens that source and reads what it actually says, which is how a misquoted number gets caught. When a claim names none, the AI searches, finds sources, and shows them. Every verdict carries its sources, and each source carries what it said.
Each source is read against the claim and marked: backs it, backs part of it, contradicts it, or says nothing. A claim is eaten only when the sources don’t back it. “Not backed” is not the same as “false”, and the checker will say which. It will get things wrong sometimes, so any eaten sentence is one click from coming back.
It doesn’t exist yet, so this is the plan, and the plan is to keep that small. It would run on your machine and look pages up by a scrambled fingerprint of the address, never the address itself. It would skip logins, banks and mail, and you could switch it off for any site.
It will, sometimes, and today’s crawler more than the checker will. Any eaten sentence is meant to be one click from coming back, with the reason it was eaten beside it. The crawler only changes what you see on your own screen, never the page itself.
The plan is that its creator fees pay for the checking. It is not a promise of anything, least of all a price. The contract address will be on this page when there is one, and nowhere else is real.
Any public page. Some sites block it, and very long pages are cut off. You can also paste the text itself.