> > If you're not running the code, what advantage do you have 
> in analysis 
> > if the objects are "real" classes/methods etc as opposed to other 
> > (package) objects modelling them?
> 
> Joshua made some good points.  I will just add that loading 
> the actual objects is conceptually cleaner than loading them 
> into psuedo objects, and avoids redundant behavior and the 
> maintenance between them.

I think that it is ulimately better to have one model, not two separate
ones (one which is used for reflection, the other for running).
However, as we found with 3.3a, if you don't have a useable system while
you're building a new one, people won't tolerate it.  For that reason, I
think a good strategy is to have two separate models (the existing one
and the new one) as the system is being developed.  When the the new
model is sufficiently robust and tested, then you can dispense with the
old.

- Stephen