the real trick is to do the debugging/exploration on a small subset of the data. Then usually you don't need all these extra measures because the real processing is only done a small number of times.