nice post, and I agree 100% with what you said. I would just like to add that, if possible, one should begin by running standard tests from the field of application of the code for which correct results are known (e.g. shock tube tests in hydrodynamics). Then one could design more elaborate tests if necessary.