What it does
It matched TexasSolver's top action in 63% of 30 spots. In one trap, it kept overbetting the nuts until the prompt explicitly named the opponent's hand rank.
Benchmarks & research
See how Jev fares in solved poker spots—and how much the wording of the game state changes its play.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It matched TexasSolver's top action in 63% of 30 spots. In one trap, it kept overbetting the nuts until the prompt explicitly named the opponent's hand rank.
Before writing code, gather game situations with known correct plays to use as your test key. Have a developer calculate numbers like pot sizes and hand rankings in regular code first. Your app can then feed these plain facts to Jev to test its choices.
Do not trust Jev to figure out who is winning on its own. In the experiment, it made terrible bets until the text clearly spelled out that the other player held a better hand.