I've posted a couple questions out at:
Here's a condensed version, with some additions:
1. Should a WSDL be produced for the ultimate sender? If so, should it be stored in the registry? If not, why not?
2. If "Contract First" is considered a best practice, should it applied to ultimate senders in addition to ultimate receivers?
3. Should a WSDL editor create two WSDL's at once (one being the invocation document advertised by the sender, the other being a service-side WSDL)?
4. Imagine that you have two WSDL's: one representing the ultimate sender, the other representing the ultimate receiver. The outbound interface on WSDL #1 matches the inbound interface on WSDL #2 (an operation signature match). Should the 'component-service' composition environment visually enable a 'snap together' programming model?
5. Let's call the thing that components and services snap into the "SCB", a variation of the PCB (printed circuit board). Should the SCB dictate the flow-of-control in a proxy like fashion (like bpel)? Should the SCB act as a message router focusing on endpoint resolution?
These are important questions. As one WS-Luminary told me, "composition is the killer app". Can I be any more blunt? ;-)
Ultimate Sender = the caller, the consumer, the client, etc.
Ultimate Reciever = the service, the producer, the recipient of the call, etc.
Contract First = the practice of stubbing out functional interfaces and non-functional policies prior to writing the implementation.