July 1, 2026

Inspecting a Gate, Live

A short walkthrough of live gate inspection during the workshop: what teachers see on screen, what they do not, and how the exercise teaches epistemic humility (Class C, Class E).

Halfway through the workshop, we stop the demonstration mid-run and click a single yellow box on the gate strip. Everything the model was about to do freezes on screen, and we read together, out loud, what the gate actually saw.

The setup

The exercise uses the same workbench teachers meet earlier in the day, unchanged. A small generative model is producing a prediction over a short passage, in the sense described by the active-inference literature: a set of beliefs about hidden causes that generates what should be observed next, then updates itself against what actually arrives (Class E, following Parr, Pezzulo, and Friston, 2022, Active Inference, MIT Press). We are not adding anything new for the demonstration. We are slowing the run down so a room full of teachers can watch one gate fire, in real time, and inspect it before the next step runs.

Before we start, we agree on a rule out loud. We will not guess what the gate is doing. We will click on it, read the reason string, look at the numbers, and then form a sentence about what happened. That rule is the lesson. The workbench is set up so that rule is possible to keep (Class C, workshop configuration).

What teachers see

When the gate fires, four things appear on the pane.

One is the gate's name and a plain-language reason string. This is written by whoever authored the gate, in ordinary sentences. In a well-authored gate the reason string reads like a note a colleague would leave you, not like a stack trace.

Two is the pair of numbers the gate compared. On a model gate this is usually a prediction and an observation, with the divergence between them shown as a readable value (Class C). We do not hide the number behind a color. The color is a summary of the number, nothing more.

Three is the version tag of the model that produced the prediction. This matters more than teachers expect at a glance. A gate's meaning depends on which model produced the belief the gate is checking, and the workbench refuses to show one without the other.

Four is the last few updates that touched the belief in question. This is a short scroll of small events, each with its own timestamp and its own reason. A teacher can walk backward through them and see, step by step, how the belief arrived at the value the gate just objected to.

What teachers do not see

The exercise is also a lesson in what the workbench does not show, on purpose.

Teachers do not see a confidence score presented as a single headline number. We removed that early. It invited a kind of trust that the underlying system does not earn.

Teachers do not see a natural-language explanation generated after the fact by a separate model. The reason string is the one the gate author wrote. If it is unclear, the fix is to rewrite the gate, not to layer a second system on top of it.

Teachers do not see a claim that the gate proves the model is correct. A gate that passes is a gate that did not object on this run, on this input, at this version. That is a smaller and more honest statement than "the model works."

Why this teaches epistemic humility

The exercise teaches epistemic humility because it puts a clear boundary around what a teacher just learned. After clicking through one live gate, a teacher can say precisely what they now know: the gate saw these two numbers, wrote this reason, and let the run continue, or stopped it. They can also say precisely what they do not know: whether the model would behave the same way on a different passage, at a different version, with a different learner in the room.

That gap between the two sentences is the whole point. It is the same gap we keep in our public stance. 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. Inspect the gate, read the reason string, and help us find where the workbench misleads a teacher.

Where to go next

EvidenceECTagsgate-inspectionactive-inferenceclassroom-practiceevidence-classesworkshopepistemic-humility

Next steps

Bring this into a working session.

The Workshop is where these notes turn into receipts on real classroom work. The Mission page is where the underlying framing is laid out in full, with the falsifiers attached.