Having used OWL-S for some time, both in the form of the `Virtual Machine' and our own experimental implementation, the Cashew project has drawn a number of conclusions about its existing process model which we assert as follow:

1. OWL-S is not service composition;
2. OWL-S defines orchestration but not choreography;
3. OWL-S is incompatible with choreography;
4. OWL-S is insufficiently composable;
5. OWL-S needs compositional semantics.

This position paper will details, and attempt to justify, these assertions and sketch the Cashew approach to their solution.