Comment by zambelli

Retry nudges do generate an extra LLM call, and those average extra calls time impacts are captured in the eval data.

But that's the difference between the call failing and succeeding (eventually).

On successful calls the presence of forge should be unnoticeable.