> But my informant could elucidate neither (a) why a naming 
> rule for files should be applied to elements in SGML, nor (b) 
> why IBM even had this naming rules for files.

If it's related to IBM file names (wash my mouth, I mean dataset names), I
would have expected to see the length of an ID restricted to 8 characters!

I seem to recall hearing some other theory about non-name IDs becoming
ambiguous if you adopted one of SGML's many ways of abbreviating tags - but
that's only a vague memory.

Regards,

Michael Kay
http://www.saxonica.com/
http://twitter.com/michaelhkay