The workbench is a piece of glass with three panes. You can point at any number on it and ask, "where did that come from," and the workbench will show you. That is the whole promise, and the whole reason we teach it.
What you are looking at
The center pane shows a small generative model at work. This is the technical vocabulary teachers will meet in the workshop, and we use it plainly here: a generative model in the active-inference sense is a set of beliefs about hidden causes that predicts what should be observed next, then updates itself against what actually arrives (Class E, following Parr, Pezzulo, and Friston, 2022, Active Inference, MIT Press). On screen this looks like a small graph of nodes, a running prediction, and an update trace. Nothing is hidden behind the graphics. Every node has a click-through to the equation and to the last few updates that changed it.
The left pane is the input. In the workshop we feed it the sort of signal a teacher already knows how to read: a short passage of text, a classroom prompt, a piece of student-adjacent synthetic data. No real student data is required for any of the demonstrations, and none is used (Class C, workshop configuration).
The right pane is the gate panel. Each gate is a small named check that either passes or fails, with a reason string attached. Gates are how a running model tells you what it thinks it is doing, and they are how a teacher tells the model what it is not allowed to do. We treat gates as the honest surface of the system. If you cannot see the gate, you cannot trust the answer.
Where the gates live
There are three kinds of gates on the workbench, and the workshop teaches you to spot all three.
Input gates sit on the left pane. They decide whether the input is well-formed and inside the scope of the current lesson. A refusal here shows up as a red gate with a plain-language reason, not as a silent shrug.
Model gates sit on the center pane. These are the interesting ones for a teacher. They check whether the generative model's current belief is still consistent with the last update, and whether the update itself was small enough to be reasonable. When a model gate fires, the workbench pauses the run and shows the divergence between prediction and observation as a readable number, not as a black box (Class C).
Output gates sit on the right pane. They check whether what the model is about to say is inside the space the teacher has declared safe for this lesson. In a classroom the output gate is the one a teacher will edit most often. The workshop shows you how to write one in plain English and how to read the pass/fail trace it leaves behind.
How to read a run
A single run leaves a strip of small labeled boxes across the bottom of the screen. Green is a pass, yellow is a pass with a note, red is a stop. You can click any box and see the exact reason string the gate wrote, the numbers it saw, and the version of the model that produced it. In the workshop we practice reading these strips together until the strip itself becomes the lesson plan for the next class.
What the workbench does not do
The workbench does not diagnose students. It does not score essays. It does not replace a teacher's judgment about a child in a room. It is a piece of instrumentation for the model, not an instrument aimed at people. When someone asks whether the workbench "understands" a student, the honest answer is that it does not, and the workbench's gates are what let a teacher demonstrate that clearly.
The workbench is also not a claim of achieved general intelligence. Our stance on this is public: 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.
Where to go next
- Come see it running: the workshop.
- Read why we teach the model, not the tool wrapper: teach UNI, not LLM tooling.
- Read how gates become curriculum: curriculum gates that teach thinking.