When results look noisy, the instinct is often to add more variables. I have had better outcomes by doing the opposite first.
My sequence is simple:
- lock a baseline route and repeat it enough to quantify natural spread,
- introduce one controlled perturbation,
- compare effect size against baseline spread before expanding scope.
This keeps decisions tied to measured signal instead of optimistic interpretation.
It also makes collaboration easier because every change has a clear reference point.