The gate

There is one rule in this curriculum you never cut. A learner does not hand work to an AI until they can say how they would know it came back wrong.

That is the gate. It sits on the third column of the board, it is the fourth line on the card, and it is the question you ask more than any other while a session is running. Everything else here can be compressed, reordered, or dropped when you lose a week to a fire drill. This cannot.

Why it holds

Generative AI is fluent whether or not it is correct. A learner who cannot evaluate the output cannot tell those two states apart, and fluency reads as correctness to almost everybody the first few times.

So the ability to check is what licenses the handoff. Not enthusiasm, not familiarity with the tool, and not how good the prompt sounds. If a learner cannot describe what a wrong answer would look like, they are not delegating the work. They are just receiving it.

This is why the gate is a knowledge question wearing a process question's clothes. "How would you know?" is really asking whether the learner understands the thing well enough to supervise it. Often they do not yet, and that is fine. It tells them what to go find out, which is what the middle column of the board is for.

What it sounds like

A real one. "I'd check that all three versions still say mitochondria, because it's on the test." Specific, and you could run the check yourself. Open the gate.

A procedural one. "I'd read it." Reading is not checking. Ask what they would be reading for.

A deferred one. "I'd ask you." Sometimes legitimate, usually a stall. Ask what they would want you to look at.

An empty one. "I'll look at it and see if it's good." This is the one to listen for. It passes if you have never had to answer the question yourself, and it fails on contact if you have.

The fourth answer is not a discipline problem and it is not a sign the learner is coasting. It usually means they do not yet know enough about the subject to have an opinion about the output, which is exactly what the gate exists to surface.

Holding it when the session is late

The gate is the first thing facilitators drop when time runs short, because skipping it feels like efficiency. Twenty minutes left, everyone is building, and stopping to ask five people how they would check feels like an interruption of the work.

It is the work. A session that produces finished artifacts nobody can evaluate has taught the opposite of what it set out to teach, and it will look like a success from the back of the room.

If you are short on time, the drill can drop to two questions and the ladder discussion to a single example. The partner test can go. Keep the gate.

One question, asked out loud

Point at one item a learner has written in the third column and ask how they would know it came back wrong. That is the whole check. It takes about fifteen seconds, it works in a classroom and a library equally well, and you can ask it without understanding anything on their screen.

The learner is the one who has to answer it. You only have to hear whether the answer is real.