Philosophers have discovered that classification systems generally depend on one's viewpoint, and are generally not universal. Thus, one should not hard-wire them into programming code, but instead make classifications into meta-data of some kind so that viewpoints can be changed or added to without large impact to the code.
With a format such as XML, the data seems to be wrapped inside classification tags with things like attributes that classify what that data is. Putting the classifications into data can make it hard to distinguish what is the data and what is not the data. Many people confuse text formats (the underlying storage) with a human model. The human should not care what storage system is used for the classifications - and XML forces one to see the classifications as data itself. BrainDamage.
See LimitsOfHierarchies, EverythingIsRelative, PayrollExample, EmployeeTypes, AutomaticClassification
JargonFile: GreenBytes? [http://info.astrian.net/jargon/terms/g/green_bytes.html] (antonym)
CategorySemanticWeb to collect this thought somewhere. --JohnFletcher