Every model should be informative enough to carry high cognitive value, and simple enough to be operationally feasible. But what criteria should one use to balance these in research? Is there any difference between formal and empirical sciences on this?