The value of a control group

When you change a trading rule, do you compare the new version against the original using the same historical period? How do you know the improvement was caused by the change?