JevMade hello@JevMade.com
← Back to experiments

Benchmarks & research

jev-sonar

Battleship where Jev returns a probability of a ship for every open cell, creating both the heat map and the shot.

Source screenshot of jev-sonar
SOURCE SCREENSHOTFull screenshot ↗

What it does

Those cell probabilities are scored with Brier loss against a seeded opponent. The headline live figures come from a single game, while automated checks use an offline oracle.

How you can use it

Your developer can read the Python files to see how the game checks a whole board. You could adapt this approach for your own strategy games. The app asks an outside AI service called TypeSafe about every open space. The service returns a simple number for each spot. The highest score then becomes the next move.

The creator shares this project under an open license. You can borrow the logic to check many map locations in a single turn. To play the live game, you must get an access key from TypeSafe. This secret text string connects the software to their system.

Maker-reported (not independently measured by JevMade): 101 questions answered in 0.454 s for the empty board, measured live against jev-1.13.0 (author-reported) · One live game: 54 requests, 4,023 questions, average 215 ms per request, per-shot Brier 0.1092 (author-reported) · Offline 20-game seed-11 run: 75% win rate (15W/5L), average 44.1 Jev shots, per-shot Brier 0.1615 (author-reported)

Primitives
noul, choice
Platform
Python
Added
Project created
GitHub stars
0 (snapshot, not live)