Half of scientific honesty is what you refuse to say. This is our standing list of claims we deliberately do not make, so a reader can hold us to the same list tomorrow.
Why write down the non-claims
A research program is judged as much by the language it declines to use as by the language it publishes. Marketing pressure moves in one direction: it asks for stronger words, sooner. Our position (Class E) is that once the audience cannot tell where the evidence ends and the pitch begins, the program has already lost the reader it wanted, the careful one. Writing the red lines down in one place, and citing them from every claim page, is how we keep the gap between what is observed and what is asserted honest across sessions and across authors.
Claims about intelligence
We do not say we have general intelligence. We do not use the artificial general intelligence label for our work. We do not describe UNI as a general intelligence system that has arrived. What we do say is that UNI is a working hypothesis on an attainable path toward General Natural Intelligence: a natural, active-inference approach whose evidence is growing, evidence-classed, and tested in the open. The wording is comma based on purpose: path is movement, not arrival, and natural marks the substrate we are modeling (Class E). The framing is defended at length in why we refuse the AGI word.
Claims about external endorsement
We do not say the preprint has been checked against Parr, Pezzulo, and Friston (2022) as a settled mapping unless the exact equation-to-example mapping is published on the same page as the claim (Class C). We do not describe UNI as endorsed by any researcher who has not written that endorsement themselves. We do not describe UNI as an official partner of any external program, or as recognized inside any external pathway, without a signed public statement from the counterparty. The rule is symmetric: we cite others accurately, and we let others describe their own relationship to our work in their own voice.
Claims about clinical outcomes
We do not offer therapy, treatment, or diagnosis. We do not say UNI heals, lowers, or treats trauma, and we do not describe the labs as brain rewiring. The active-inference labs are a modeling and design lens for computational phenotypes, hypotheses at the algorithmic level, not diagnoses at the person level (Class E). UNI is not a clinical tool and nothing on this site is medical advice. Where the labs touch human behavior, the labels are candidates for the modeling community to argue about, and they arrive without a promise attached.
Claims about accessibility
We do not say active inference is easy, that anyone can do it, or that no technical background is needed. Building a POMDP generative model, choosing a factorization, and reasoning about variational free energy and KL divergence between the recognition density and the posterior is technical work, and pretending otherwise sets a reader up to feel stupid when the math shows up. What we do commit to is a working glossary (Class C), plain-English labels next to each equation on the labs, and a reading path from the free energy principle in one sitting through the ELBO and expected free energy posts.
Claims about arrival
We do not say the preprint is peer reviewed while it is a Zenodo preprint, and we mark the preprint status on the science page every time we link it (Class C). We do not say the Stratified Palimpsest benchmark has settled a debate. It is a pre-registered falsification instrument, and it has published families where a UNI controller loses to a neural baseline. Loss cases are published on the same page as win cases, because a benchmark that only shows the wins is a demo.
Language we keep off the copy
A short list of words we do not use for our own work: solved, achieved, breakthrough, revolutionary, the only, endorsed (unless we can link the endorsement), peer reviewed (unless the paper actually is). If we ever use one of these, treat it as a copy error and please tell us. The list is boring on purpose. Boring language is what keeps a claim page readable a year later.
The commitment
UNI is a working hypothesis on an attainable path toward General Natural Intelligence: a natural, active-inference approach whose evidence is growing, evidence-classed, and tested in the open. Do not take the claim on faith. Test the build, inspect the gates, and help us find where it fails. This non-claims list is one of the gates. If any page on the site crosses it, that page is wrong and we will fix it.