Maybe what's needed is looser, first-pass categorization. Then increasing waves of finer grained categorization.

Of course, this iterative approach supports the person trying to do the categorization, but how useful is it for someone trying to use the categorization? Maybe "a bit" useful to begin with, then more so. If the user / user-software can handle the degrees of approximation.

Compare :

See also :

  • ExtremeProgramming (another iterative approach to design, this time based on continuous ReFactoring) Any categorization parallels of SpikeSolutions?