01
Set in advance
Every test is published before kickoff and cannot be changed later.
HOW IT WORKS
A new model takes over only when a separate test proves it is better. Failed attempts remain visible.
Every test is published before kickoff and cannot be changed later.
Wins and misses automatically count toward the same public score.
The AI finds weak areas and cautiously tests new variants.
A new model takes over only after an independent test confirms the improvement.