Tips & tricks

Practical levers for a clean, fair pilot.

Normalize bounds and measured statistics before the race. Keep the same starting conditions, random seeds, and measurement budget for every contender.

Use per-channel feedback whenever the instrument exposes it. Collapsing a vector into a scalar destroys information the loop can use.

Warm-start related tasks, log every configuration and measured output, and compare steady-state error, not only the first few updates.