What it is
A browser-based harness where Laya (local, open source) and Jev (TypeSafe API) play Snake and a Mortal-Kombat-style fight. Every move is a real model decision, not scripted.
How it uses Jev
Jev answers typed questions per game: in Snake, a choice over steering directions and a noul on sprinting; in Kombat, a choice over actions and a noul on committing to a heavy attack. The server proxies Jev calls, and the game uses the returned answers to drive the agent.
Primitives:choicenoul
Technique worth stealing
Ask both models the same typed questions on identical state, and pace the game to the slower model's latency.
Try it
docker run -d -p 80:8732 -e TYPESAFE_API_KEY=your_key_here ghcr.io/promptengineer48/laya-vs-jev-arena:latest
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.61
- Shows a System One patternno0.34
- Handles uncertaintyno0.03
- Measuredno0.05
- Runnableno0.19
- Worth recommendingno0.19
- Model replicano0.03
- Problem scopescore on a 0–2 scale0.00
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 22 Sept 2026.