When building XML systems, how do you choose the best granularity for storing and retrieving fragments? 

Machine to machine

Human to machine

Human to human

Part of the art is interpreting what branch and leaf combinations best give a role/user the most copacetic view.  How do you choose?  Does the user choose?

The proportion of XML consumed and emitted by machines or humans is not interesting,IME.   The cost and type of the value-add of the humans consuming and emitting XML is.  In documents, this is obvious.   Granularity.

len