The practical mistake is to respond with another broad experimental round instead of identifying which constraint changed regime first. If the decision is still ambiguous, more runs only add cost.
A useful diagnostic sequence starts with three questions: what changed between scales, which variables were implicitly held constant in the original interpretation, and what small experiment set can falsify the leading competing explanations fastest.
For many teams, the best next step is not a large redesign. It is a ranked memo that narrows the mechanism space and a minimum experiment set tied to the actual business decision.