Designing a Trading Experiment

If you wanted to test whether a particular chart feature improves entries, how would you structure a fair experiment?