The five labs
Move the dials yourself. Then try to break them.
Five small, interactive labs, each one a window onto active inference. We tell you plainly what each one shows, and where the metaphor ends. They are tools to think with, not clinical instruments. The honest part is the boundary, not the demo.
Precision Lab
What it shows
An agent feels its way through a maze using only the walls it can touch, and one dial (trust in its senses) decides whether it finds the goal or wanders.
Where the metaphor ends
The dials are properties of a toy agent in a toy maze. They are not a brain state, a diagnosis, or a person. A resemblance to a behavior is an interpretation, not a measurement.
Echo Lab
What it shows
The same agent, the same maze, but now it listens to echoes that reach two cells out. You can watch how a longer-range sensor reshapes the very same inference loop.
Where the metaphor ends
Same fence as Precision: nothing here is pharmacological, diagnostic, or therapeutic. The maze agent is not a person, and a dial is not a brain state.
Loop Lab
What it shows
A two-state model of a stuck pattern. You can watch standard exposure stall, and then watch a grounding-first sequence succeed, with the same model and the same starting belief. Only the order of steps changes.
Where the metaphor ends
People are not two-state systems. The behavioral words (avoidance, exposure, grounding) are vocabulary from the literature applied as analogy. It is not a diagnosis, not a treatment plan, and not validated against any clinical outcome.
Heart Lab
What it shows
A reduced long-term cardiovascular and kidney loop settling into a steady state. Its point: a disease state is the loop regulating to the wrong set-point, not the loop breaking.
Where the metaphor ends
A teaching model, not medical advice. It models long-term blood-pressure and fluid regulation, not an acute event. Every reading is an interpretation of the math, not a measurement of any patient.
Cell Lab
What it shows
A self-managing digital service cell hit by hidden disturbances. A UNI active-inference controller, judged against rule-based and random baselines, and built so it can be proven wrong. A loss is reported as plainly as a win.
Where the metaphor ends
Pre-registered and reproducible, not operational advice and not a consciousness or AGI claim. Autopoiesis here means one thing only: keeping the cell inside its operating band.
These run live in your browser on the UNI site. The math is the same active inference described in the preprint, an unrefereed working paper with expert review pending. Open them, move the dials, and watch what changes.
How the harder work is judged
The Lab Protocol: one cure at a time, gates written first.
Beyond the browser labs there is a research discipline we hold ourselves to. Every change is tested against a matched control where one variable moves. Every test names its pass line and its falsifier before the run, not after. Verdicts are PASS, PARTIAL, FAIL, or WITHHELD, never a percentage and never spun. And every claim points to a commit, a saved artifact, and a log that reproduces it.
Pre-registered gates
An adversarial review
Verdicts that can lose
Operational behavioural and organisational measures are necessary-not-sufficient substrates with ZERO evidential weight for awareness, consciousness, or life on their own. Passing a gate demonstrates the named behaviour, never experience.
Built to be proven wrong
The Cell Lab is a pre-registered benchmark.
Its claims and their falsification criteria were written down before the runs, so the result can go against us and still be reported in full. You can upload your own controller and run it head to head. If a baseline beats UNI, that is the result, shown plainly.
The intelligence underneath the work.
These labs make the science watchable. The same thinking is what we carry into your engagement.