With release 2, it’s noticeable that the developers have grown more accustomed to the ICONIX modeling process, and in fact they’re starting to drive new features from the use cases and sequence diagrams. So (perhaps for the first time in recorded history?) we actually accomplished the feat of having the model and the code converge over time, as opposed to diverging over time. (Historically, models and code always have tended to diverge over time. Reverse-engineering tools have helped a little bit back at the class level, but nothing has really had much impact at the use case level.)

Perhaps another way to state this is that a feedback-driven process also has benefits in terms of improved modeling technique, with a short first release taken all the way through to code and then feedback on how the models could have been improved. The developers, seeing the resulting improvements in the code definitely took the modeling more seriously in the next iteration and avoided a lot of potential rework on both the models and the code. (Restated once again:“Eureka—it works!”)

The issue of code/design divergence is discussed further in Figure 7-1).