This time I want to talk about the “next” step in the requirements journey. I can right away say that I am a fan of agile and rapid development, but for the sake of this chat I like to describe a stepped approach. A newly formulated requirement should materialise on the solution side as one or many design elements. These elements form part of the architecture.
If you remember the blog here by Hillary Sillitto he described different views: System, Operational, Physical, Functional. The book he have written on architecting is brilliant and really to recommend. All of these views should capture different aspects of the design suggestion that eventually satisfy all the stated requirements. That’s all well, but what tend to happened in may cases?
We really now start to compare apples with pears (or should I say fruit salad?). Its really hard to compare written text with models. Are they correct, complete or are we missing something.
So, is this a big deal? Well if we make a recap from last post. I then argued that most errors (and I then mean errors not found with a spell checker) are brought to the architects and system designer I need to validate that their proposed solution is Ok, right? But If I can not really penetrate their ideas because they are using other techniques, tools and languages I might loose that ability. Do you agree?
The graph really show that the architects and designers are pretty good at doing their job and they find more errors due to the bad requirements than they produce and that’s a relief, but they will probably not find those errors that are caused by ambiguities, missing parts or plain wrong information. The architects produces and architecture that is supposed to cover all aspects that has been stated in the requirements, but they should not do anything if they don’t find a requirement justifying the design element or function.
One can argue that techniques like Model Based Systems Engineering (MBSE) are solving a great number of complexity issues and you can do so much magic when you are able to look at those views to see the complete picture, but you can not really validate it with people that don’t “read” SySML. Those people are all the stakeholders, customers, users that you so desperately need to communicate with and they don’t speak your language as an architect. Ok, again its now possible to add a requirement view in the modelling tools, but then you are just adding a notepad functionality and you miss all of issues of having correct semantics and syntax. You are basically back where you started (you have just moved them to another domain and probably tool).
Major problems in the transformation from in this case written requirements in one language (let’s say English) to an architecting language like UML or SYSML is the inability to validate Correctness, Completeness and Consistency. A rather academic picture and text I have stolen from my friend and colleague Prof. Juan llorens (sorry!) is this:
Bearing in mind that the goal of requirements is to specify a product, service or system, then we have two complementary points of view to define good or bad requirements. From the point of view of the sender, the final quality of requirements is validability, i.e., the sender must be able to confirm that the system architecture effectively expresses the system solution that answers his or her needs. This property can be further unfolded, in a second level, into three other properties: completeness, consistency and understandability.
From the point of view of the receiver, instead, the two essential qualities are: verifiability, i.e. it is possible to check that the developed system architecture meets the specified system requirements; and modifiability, i.e. it is possible to modify the architecture to enable change. Verifiability depends on the same three second-order properties already mentioned (completeness, consistency and understandability), as well as on three other properties: unambiguity, traceability and abstraction; modifiability depends also closely on these three last properties.
Unambiguity and understandability are interrelated (according to some, they would be even the same property), since, if a requirement is ambiguous, then it cannot be properly understood. Note also that abstraction and unambiguity are not opposing concepts: a requirement must have a single interpretation at the abstract level of specification (unambiguity) and at the same time must avoid technical details about the implementation (abstraction).
Other things worth mentioning is also that organisations seldom have one tool from one tool vendor, they have probably multiple tools spread and used differently in different parts of the organisation and to that we can ad a layer of self-made hacks, spread sheets and databases.
At one point I will write a future blog post on using natural language processing, requirement patterns and techniques to decode the language so that one can create requirements out of models (or vice versa), but for now I will say good night, auf wiedersehen and hope you read next blog on the test phase.
I have called it “the test catastrophe, but lest push on for that delivery”.