Every time we touch a dataset we are told "in this case x means y else x means z only if x = 999 or else z means -z if.....". Or in plain english a pink cat is a dog unless its not alive (as indicated by its age) then its a toy and its age is -999.
I was told by one progammer the reason for such convolutions in data and he does have his point. "I write the smallest, easiest to maintain most efficient to run code..... I am not writing extra code to make sure the data is clean and meaningful" (I paraphrased).
