Carlos comments on the one-sided nature of Liskov's Substitition Principle.
As I've said many times, in a document-centric, message-centric world, objects are not the center of the universe so API substitutability is not a useful metric for compatibility. (Example, in REST, there is only one "API" so substitutability does not arise).
I think Carlos is right when he says we need to look at both the "provider" and "requirer" sides. We need to focus on foreward and backward compatability of message *notations* by means of a message processing model, not API comptability through polymorphism.
RIG 6 in the interoperability guidelines I am working on with the Irish government for the public services broker, addresses this issue.
Dave Orchard is well worth reading on this topic.