Video unavailable. Open the source below.
NewHow it uses Jev
For each guard, the server sends that guard’s local context and world alert alongside typed questions about threat (Noul), suspicion (Score), tactical intent (Choice), and attention target (Choice). Deterministic server code validates freshness and legal actions before applying proposals to the simulation.
What Jev decides
Current threat checkExample answers · not a recorded Jev response · Source ↗
Question 1 · threat
YOUR APP
INSTRUCTION
Using only this guard's context and the world alert, is there current evidence of an intruder or active security threat?
STATE
World tick and alert, plus this guard's local context from the decision batch.
JEV · NOUL
YesNo
App workflow
- Simulate the world
The authoritative server advances the game and builds local observations for each guard.
- Request guard judgments
A batched Jev request asks four typed questions per guard from the supplied context.
- Validate proposals
The server checks response types, context freshness, legal targets, and policy constraints.
- Apply and inspect
Accepted intent affects the simulation; the Decision Lens and JSONL traces expose the applied judgment and evidence.
Why it is interesting
The game exposes the evidence and full applied judgments behind guard behavior while keeping physics, navigation, legality, and world outcomes under deterministic control.