What problem does it solve? Building a playable grid world for the Jev agent requires a valid arena.json and a way to confirm the world is actually playable, which is hard to judge without live feedback on how Jev reasons through it. ## Core Features & Use Cases - Live world editing: Update arena.json and the OpenHarness viewer adapts in real time, with no restart or second server. - World verification: Run node "$JEV_DSH/toolchain/check.mjs" to validate that arena.json is a playable world and report its verdict and status. - Behavior observation: Watch Jev's per-action probability distribution and confidence for up/down/left/right/wait to see whether the arena produces both confident and confused states. - Programmatic evaluation: Use toolchain/jev.mjs to ask Jev structured questions about an arena, with a deterministic mock when TYPESAFE_API_KEY is unset. - Use Case: You design a maze layout, run the checker to confirm the goal is reachable, then watch Jev play to verify its confidence dips and recovers at interesting decision points. ## Quick Start Edit arena.json in the workspace, then ask the agent to run the arena checker and watch Jev play the updated world in the live viewer.
