Systems engineering thought leader, consultant, trainer and coach, impacting people's lives on six continents.
A client just asked me this question. Given the issues that it raises, I suspect there will be interest in my answer. I start by pointing out that there is no single answer to the question. The answer lies requirement-by-requirement and depends on the information content of the requirement, the importance of the requirement (based on the amount of loss if the requirement were not met), and the probability of the requirement not being met. The last two factors combine to constitute risk. The aim is to maximize the difference between the risk-reduction benefit of verification of satisfaction of the requirement and the cost of achieving that benefit, giving an optimum verification requirement that drives verification design (how we will verify). A verification requirement defines the characteristics or quality of evidence required that a system (etc.) requirement has been met. The optimum verification requirement is usually best decided upon by discussion by a group of three engineers experienced in both verification and development, although a rigorous method of developing verification requirements also exists. Non-functional requirements are those in sections 4.1, 4.2, and 4.5 to 4.10 of a well-structured system requirements specification. If we are working in a solely database environment, they are all of the requirements engineering types except for functional, function and performance, and state/mode transition and response requirements.
