> 2) we may rely on system engineers to deal with unsollicited in a 
>responsible way. Many servers have extensions that might be useful for 
>others. Only extensions that are not useful for others should be on 
>request only

This is my main problem with the proposal.

1.  Surely if an extension is not useful, it wouldn't exist?

2.  Assuming that there are non-useful extensions, how do you determine 
if an extension is or is not useful to the current client?

3.  Assuming that you have a well implemented server, you could return 
an AWFUL lot of extra data about a record, the search, the terms in 
scan, etc.  You're just wasting everyone's bandwidth and processing time 
by sending unwanted data that the client probably doesn't know what to 
do with.  Yes you can safely ignore it, but that is not a zero cost 
option.

4.  If, as a client, you want to get back anything, we have a way to do 
it -- the accept extension.  I don't see why this should be forced on to 
everyone when clients that want it can easily have it.  Just put it at 
the beginning along with the static operation and version parameters and 
forget about it. :)

5.  For some clients, getting back unknown elements could break them if 
they use very strict XML processing rules.  Enforcing something that is 
known to break otherwise well implemented clients is not good behaviour!

Rob