Extrapolating from your comment, I guess we'd kind of want a blocked design that is placebo-vs-caffeine and exercise-vs-none. That would tell us whether there

Extrapolating from your comment, I guess we'd kind of want a blocked design that is placebo-vs-caffeine and exercise-vs-none. That would tell us whether there is just an additive effect of high-dose caffeine and exercise, or if somehow the caffeine is interactive with exercise (i.e., has a bigger effect in the exercise group than in the rest group).