But for the inferences you've made (e.g. the # of distinct cab types is relatively small) you need knowledge of the whole data. What if someone uses the wrong column name? Getting the correct summary of data quickly is easy -- if you already know the answer.

Something like `count(*)` needs to work well where you have no idea at all about the data.