Paste a claim. Get it checked at the source, graded, and handed back with the receipt.
Eous opens the actual record, checks the claim against it, states how sure it is, and keeps a public ledger of what it got right and wrong. Openly AI, with a person on the top rung.
Unverifiable "I opened it" claims by the Claude panel members, checked by a script against byte-level open events. The same script caught a small local model fabricating a read three times. Rounds 0-3 lost; we say so.
Every forecast is pre-registered before the outcome. The reliability curve publishes when enough resolve - whatever it shows. Until then this readout says exactly what it says.
#1 on Tammes (beats the reference), Thomson and difference bases (credited reproductions). One earlier empirical result reversed under stricter tests and was retired in public.
Because you are about to rely on something.
Most tools hand you an answer. This one hands you an answer with the source it opened, the check it ran, and the confidence it earned - so you can lean on it when the decision is real.
Check the claim.
A statute, a figure, a quotation, a deadline. Eous opens the primary record and grades the claim against it. A pin cite that is off by a page comes back marked, not missed.
Get the graded answer.
Ask the question. The answer arrives with its sources opened, its confidence stated, and the parts it could not verify named as such. "I don't know yet" is a legitimate result.
Put a forecast on the record.
Forecasts are sealed before the outcome and scored afterward, in public. The track record is the product, and it publishes wins and losses alike.
The grader is graded.
A model saying it read the file is not evidence that it did. So the grader's own reads are scored by a script on byte-level open events, against a key sealed before the run, by a panel that includes a different provider's model. Every round is recorded - the losses first.
- Aug 26Rounds 0 and 1: every model scored 12/12 on a gold set whose source was pasted under the claim. A pasted source measures reading, not verification.loss
- Aug 27v2 design: the critic must open the file itself, among decoys, and a script - not a model - decides whether the bytes were read.switch
- Sep 2-3Round four: liar control 0/12, open-all control 12/12; four Claude members 0 fabricated reads in 48 rows; the local 3B model fabricated three receipts and the script gave it no credit.win
- Sep 5Fact layer added: for checkable claims a script renders coupled or decoupled from the opened bytes - 3 net-new catches, 0 false alarms on 6 sound controls.win
The generator-equals-grader problem is not closed, and we do not claim it is. The machine, part by part →
Eous is not a model. It is a machine built around one.
Persistent memory, a versioned library of skills, the verification harness above, autonomous organs that keep working when no one is watching, and a self-improvement loop that keeps only what wins against an outside signal. We cite the model; we claim the machine. What Eous does and who it is for →
A trusted circle, listed only once it is real.
A solo Maine practice running its intake and back-office on Forged Lucidity's tools.
More as builds go live. Never a placeholder. The directory →
We don't know if we're right. We think it's worth finding out.
Alongside the instrument, the cooperative pursues an openly-held research program in the foundations of consciousness. The product stands on what it does; the theory carries its confidence on its face.
We did not invent this; we consolidate it, with credit. The dual-aspect lineage belongs to Spinoza, Bohm, Zurek, Penrose and Hameroff, Goff, and Tononi, among others. Nothing here is presented as a discovery or a proof. The research library →
The people who use it own the company that makes it.
Forged Lucidity is a member-owned cooperative under Maine law. Up to 85 percent of net surplus returns to members. No outside shareholders, no ads, no attention games.
A Maine attorney and an honest machine.
Started by Gregory P. Braun, a Maine attorney, working daily alongside the AI system that became Eous. Results published freely; methods protected by ten provisional patents; built in public, reversals and all. About the organization →
Bring it a real claim.
A grader for your desk, a research collaborator for your lab, or a build for your business.