A Taxonomy of Deviance?

Hi all,

there have been discussions previously here and elsewhere about such  
notions as "feasibly valid". I'm working on a discussion of the ways  
in which processors that know an XML grammar (be it XML Schema,  
RelaxNG, etc. it doesn't matter) and use it for a specific task can  
be resilient to errors in various manners and to various degrees.

I'm finding it difficult however to come up with a way of measuring  
their resilience, and was wondering if anyone had come up with a  
classification for grammar deviations and a metric for just how  
deviant an instance (or subtree) is. Any thoughts or suggestions there?

-- 
Robin Berjon
    Senior Research Scientist
    Expway, http://expway.com/