> I can't seem to recall that discussion. Can someone summarize 
> it?  We spent
> years designing Z39.50 diagnostics so that clients would do something
> intelligent with them, before completely repudiating the idea.  What's
> changed?

I don't recall that discussion either. I'm pretty sure I'd have argued
that if details were meant to be machine readable then its type should
be xsd:ANY so that the diagnostic can define its structure as an XML
structure rather than a formatted string (if I'd noticed that
discussion).

Matthew