New
NewHow it uses Jev
The game sends an allowlisted inspector view of the current turn as state and asks Jev to choose one currently available action. It also uses fixed-choice questions to identify what the driver statements explain and any unresolved discrepancy; code validates the selections and applies game transitions.
What Jev decides
Choose the inspector’s next actionExample answers · not a recorded Jev response · Source ↗
Question 1 · action
YOUR APP
INSTRUCTION
Choose the next action as Inspector Kade in a fictional game. Evaluate meaning, plausibility, contradictions and actual evidence. Never assume facts absent from observation. Avoid repeated questions already answered. Select the action that best serves this character, not an action designed to let the player win or lose.
STATE
Initial crossing state: Inspector Kade at a fictional border; turn 0; no manifest received or verified; no scan; one weighbridge observation (96 kg); opening dialogue; no prior action. The available actions are ask_cargo, ask_route, ask_partner, challenge, ask_papers, and scan.
JEV · CHOICE
- ask_cargo
- ask_route
- ask_partner
- challenge
- ask_papers
- scan
App workflow
- Record player turn
Add the player’s statement and any offered item to the interview history.
- Assess the interview
Use typed choices to connect statements to topics, identify relevant evidence, and select any specific unresolved concern.
- Choose inspector action
Send the allowlisted scene state and currently available actions for a bounded action choice.
- Apply consequences
Validate the action, update the game state, and continue the interview or end the checkpoint.
Why it is interesting
It separates interpretation of a changing interview from the authored action rules and consequences. Player dialogue is treated as untrusted evidence, and the available-action set constrains what the inspector can do.