RE: How to specify "coupled" attibutes

> I wonder if this assertion mechanism will enforce the content 
> model to be those expected, 

Yes, it will.

> or if it is a mechanism that 
> works like schematron, that is to say *after* applying the 
> content model ;

Depends on whether you are using the term "content model" to include the
assertions or not.

 an editor would propose to its user some 
> candidate elements for insertion and then refuse the one 
> selected thanks to the assertion ; not fair

I think it would require a rather clever editor to take assertions into
account when offering prompts to the user. But of course assisting authoring
is only one use case for schemas.

> 
> I'm convinced that it's much more efficient to act on the 
> content model directly ;

I'm sorry, what is the "it" that is more efficient?
> 
> Notice that mixing a declarative language with imperative 
> constructs is a concept already adopted by your peers : 
> if/then/else and for-each structures are part of XPath2 ;

There's nothing remotely imperative about conditional expressions or mapping
expressions, I think you have misunderstood the language semantics. Just
because these constructs are dressed in a syntax that is familiar from
procedural languages doesn't make them procedural.

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