Is the null hypothesis always wrong unless you have insufficient statistical power? (i.e. for any 1>a>0 there always exists a sample size n that gives us p<a)? I guess this fails only if we are allowed to use theoretically defined distributions with exact known parameters.