What is Dernier Abri?

A French 3D browser defense game where an AI (JEV from TypeSafe AI) estimates from a simple textual description whether a silhouette in the night is a zombie or a survivor.

JEV provides the classification probability; the README's rule is that the model classifies and deterministic code decides. JEV returns only probabilities, and the shot passes through explicit, versioned, tested rules.

What statistical machinery does it demonstrate?

The English blurb lists Bayesian evidence fusion, Wald's sequential test, calibration, drift detection and a reusable observe → analyse → decide → respond core.

What is the headline safety claim?

Zero humans shot across thousands of simulated games, per the README, with doubt meaning no action.

Where does the project come from?

It is a Coding for AI course project from L3 Data Science & IA at aivancity, built around JEV, per the README.

What is not documented in the captured material?

The digest truncates the principles table and shows no setup commands or JEV question shapes, so how the game calls JEV is not evidenced here.

What license and status does it have?

MIT, a Python repository, 1 star, created and pushed 2026-10-16, per the digest's GitHub metadata.

Jev directory

CommunityJev Demos & Experiments

Dernier Abri

A French 3D browser defense game used as a test bench for decisions under uncertainty: JEV (TypeSafe AI) estimates from a text description whether a silhouette is a zombie or a survivor, and deterministic code decides whether to shoot.

See highlights Open github.com

Stars: 1
Language: Python
License: MIT
Reading time: 2 min
Published by: Community
Author: Pablositoo
Last verified: Oct 2026

Pablositoo/dernier-abri

User repository on GitHub

View on GitHub

Jeu 3D de défense propulsé par JEV (TypeSafe AI) : l'IA classe, le code décide. Fusion bayésienne, test de Wald, calibration, 0 humain abattu (L3 aivancity)

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github game browser python

Works forJev for Classification Jev for Decision gating

Used inJev for Gaming

Languages: Python
Platforms: Browser
Capabilities: Open source

Highlights

JEV reads a textual description of a silhouette and returns the probability it is a zombie or a survivor; code decides the shot.
The model classifies, the code decides: shots pass through explicit, versioned and tested rules, not model output.
Built as a decision test bench: Bayesian evidence fusion, Wald's sequential test, calibration and drift detection.
Reports zero humans shot across thousands of simulated games, with a reusable observe → analyse → decide → respond core.
A Coding for AI course project (L3 Data Science & IA, aivancity); Python, MIT, created and pushed 2026-10-16.

Watch out

README and GitHub metadata reviewed 2026-10-17; the README is in French and the digest truncates the principles table at the 'doubt means no action' row, so the demo setup and JEV prompt details are not captured. All numbers are the author's own.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

X: Text-to-screenshot (Malay Vasa)

X: Google's MediaPipe Decision Maker, tested in the browser

X: Playwright CLI + Jev vs Playwright MCP

X: jev-ultrafast splits every browser decision into operation, target and value. JEV passed 18,000 stars in its…

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.

Jev Demos & Experiments#gaming#agents#control

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai