Working Group Process

To make this both tractable and fun in a working group setting, we treated the exercise as a chance to discover and practice our domain modeling skills in collaborative learning work. For example, the initial proposed scope assumed that representation issues could be usefully dealt with separately from domain engineering process issues. Testing this hypothesis required us to negotiate boundary issues, explore alternative framework structures, etc., all skills inherent in domain modeling itself.

Some of the challenges we encountered in our own modeling process were the following:

• Each representation strategy seems to embed certain assumptions about the relative dependence or independence of the various features. For example, if you model two features using a two- dimensional axis (e.g., formal expressiveness vs. closeness to system modeling notations) you get a set of quadrants, some of which may be "dead" (don't make logical sense). Our general pattern seemed to be to move towards representations that made fewer assumptions about inter-connections, and then to explore the patterns after generating some data in a given representation for various scenarios. This is how we arrived at the Kiviat diagrams.
• Beyond the representations themselves, the choices of the individual features, their order and presentation all affected the usefulness of the model. So, even with the Kiviat diagrams, we got tangled in whether temporal variability was truly separate from inter- and intra-system variability.

At the outset, several participants suggested that we could not discuss alternative approaches to domain model representation without first determining what we meant by a domain, what we meant by a domain model, and what the content of a domain model should be. While we recognized the importance of this task, we agreed to the experiment of proceeding with these as open questions.

To everyone's surprise, by starting instead from a set of examples and candidates, we were able to proceed quite well without this starting consensus. Our focus on the representation strategies indirectly led us to discover the underlying theories in use embedded in each representation strategy. Instead of arriving at a consensus, we arrived instead at a better understanding of the varying definitions of domains and domain models adopted by different communities. As a side-effect the model that explains many of the communication breakdowns that occur among researchers discussing domains, domain models and domain analysis. Armed with this it ought to be possible to manage these conversations more clearly in the future. This appears to be one of the most valuable outcomes of the working group session.

Section 11 of this report includes more detailed description of the various semantic interpretations of the term "domain" as they were elaborated in our discussion.