The dynamics involved in the training system marketplace of today are dictating the need for major changes in the way organizations specify, develop, and maintain training systems. One of the key areas affected by these changes is the system and software architecture of training systems. This is evidenced by the increased attention that has been placed on architectures by recent initiatives (e.g. Structural Model, Mod Sim, STARS, DIS, ARPA DSSA, etc.). There are many reasons for this emphasis, not the least of which is a desire to produce training systems at the least possible cost while providing faster time to market and higher quality. An architecture for training systems can be a framework to enable cost reduction, reusability, and standardization.

We derive a set of attributes which we believe characterize a "good" software architecture. We discuss an architecture developed by Boeing Defense & Space Group, the Domain Architecture for Reuse in Training Systems (DARTS) and evaluate DARTS against these criteria. We also discuss the role of DARTS in megaprogramming, part of the ARPA STARS initiative, and suggest that DARTS is a suitable architecture for achieving the STARS vision of process-driven reuse.