[ your explanation about sequence and flow and asumptions allowed ]
My point was, that with the currently specified behaviour of runtime engines, it is not
possible to model the callback scenario, as described in the 1.1 BPEL spec.
It contains a sample, where ones does invoke a invoice creation service with 2 sets of info, and the service will call back. it is modelled as a sequence at the invoker side, and this is simply not reliable. My suggestion to use a flow instead would make it a bit more reliable, but still it depends on unwritten runtime engine behaviour.
So my point is: there is a lack of sematics specification. If we accept this, we must not show examples which depend on it.