Jev is not a mind that notices the wet floor. It is an exam you author for a purpose: caution, routing, approval. A large model reconstructs the walk. A well-written Noul answers “slow down?” the way ordinary cognition already would.
Sort of. The SAT does not invent the subject. The College Board writes items, defines right and wrong, and returns a score. The student only fills the bubbles. Jev is the student who cannot write an essay. You are the board.
Fixed questions. Bounded answers. A numeric result you can set a cutoff on. Different exams for different purposes: caution, hire, publish, route, spend.
There is no official form. No national norming unless you do it. A 0.92 Noul is not an 800. It is only as good as the item you wrote and the state you fed it.
Treat one repeated decision as one exam. Do not start with a chatbot. Start with a situation that already happens every day.
State floor: wet gait: walking footwear: smooth sole task: cross the room Noul question: Should the walker slow down or change path? yes means: ordinary caution a person would take without debate Policy if noul >= 0.85 and confidence >= 0.8 → caution else → human glance
TypeSafe Playground for the first items. Then the API from a small script that saves a handoff JSON. Workers still do the walking. Jev only marks the sheet. Official start: typesafe.ai playground and the Python SDK.
“High Noul” does not mean “always yes.” It means the yes-probability is high when a typical attentive person would already act, and low when they would not. You are aiming at common sense, not cleverness.
Ten obvious yes, ten obvious no, ten messy. If obvious wet-floor-plus-walking is not a high Noul, rewrite the stem. Do not add a second model to explain it.
A Noul is not a physics engine. It will not compute friction. If you need physics, run physics in code and pass the result in as state. Jev votes on the already-seen fact.
Tell a large language model there is a wet floor and it starts reconstructing the scene: I am walking, I have a body, water reduces friction, falling is bad, therefore perhaps I should… That is System Two writing an essay about a reflex.
Re-derive walking from language. Spend tokens on where you are. May still produce a careful paragraph. Slow, expensive, and easy to distract with extra story.
State already says wet floor and walking. The Noul is “slow down?” A high yes is the score. No need to think about the fact of being a walker. That fact was posted at the door.
This is why the SAT analogy holds. The exam does not ask you to invent gravity. It asks whether this item matches the rule you already printed. The monitor is the eyes. Cortex may later write “why we rerouted.” Jev only raises the caution flag at 90-plus when the state is the state a careful person would treat as settled.
Bad item “Please consider all implications of ambulation.” Good item State: floor=wet; motion=walking; grip=low Noul: Should we slow or redirect before the next step? Yes = ordinary caution, not a research project.
Business and personal use are the same SAT with different state packets.
Freeze legal actions. Ask Jev only among those. Require code to check send, pay, delete, and post. A high Noul is a vote, not a proof the floor was dry again.