When is the Design Complete?

Abstract

Almost every software-development and software-intensive system-development methodology calls for the design to reach some level of maturity before the team moves into software and system implementation. Yet data from many sources indicate that a large percentage of software and system-development programs encounter significant difficulties. Data from my own work fixing such problem programs indicates that the major recurring theme across such problem programs is that the design was inadequate for the task at hand; this, of course, should have been detected during the design phase, before the program moved on to implementation, integration, and test. In this paper, I examine indications from a number of real programs to determine why the design is so often inadequate; one of the key findings is that our standard methods, processes, indicators, and metrics for determining if the design is complete are seriously flawed. A proposal for better design-completion indicators is provided, and the implications for practice discussed.