I do not think the current CDL specification (  Web Services Choreography
Description Language Version 1.0  W3C Working Draft 12 October 2004  
http://www.w3.org/TR/2004/WD-ws-cdl-10-20041012/  ) specifically states this,
and perhaps it should, but I assume that a Choreography Description, written
using CDL is to be interpreted element at a time in the sequence those elements
are written down following the normal rules of XML documents (lexical ordering),
except were the defined semantics of an element so encountered dictates
otherwise - particularly the parallel and choice constructs in the case of CDL
(not quite sure why we need sequence as that is the default ordering). 
   
 Following on from this idea one naturally assumes that conditionally
expressions are evaluated when (and only when) they are encountered.  This rule
would seem to apply well to the guard attribute which is evaluated when a work
unit element is encountered.  However, it does not apply if we want the same
guard condition to be re-evaluated at any later point (as the current text
implies rather confusingly).  The current text in place seems to state that the
repetition condition is also evaluated on entry to the work unit.  It is not
clear to me that this is useful and it would seem better to state (as implied
elsewhere) that the repetition condition is evaluated when the work-unit is
completed and before going on to the next element / step to see if the work unit
should be repeated.  The text should state that once a repetition condition on a
work unit has been tested and found to false (maybe after several previous
evaluations that evaluated to true) then that work unit is left behind and not
used again. 
   
 The 'complete' attribute on the choreography element does not seem to fit this
pattern at all and so it should be clearly specified  when it is evaluated (at a
particular point, at several points, continuously while the choreography is
active), and what happens when it evaluates to false, and what happens when it
evaluates to true. 
   
 Asking whether a choreography is complete as you start it seems a little odd,
and asking the same question at the point the choreography element is complete
and the description is about to end or go onto the next step seems even odder. 
So it should be clearly specified  when it is evaluated and what happens when it
is evaluated.