I think the answer is obvious so maybe you're looking for something different.
More perf means more attempts in parallel with some sort of arbiter model deciding what to pick. This can happen at the token, prompt, or agent level or all of them.
More perf means more attempts in parallel with some sort of arbiter model deciding what to pick. This can happen at the token, prompt, or agent level or all of them.
I ask it another question later, or maybe troubleshoot. It tells me B is causing the problem, and I should do A instead.