Why do we ignore the quality of traceability?
Traceability is undoubtedly an important concept in systems engineering. In many cases, traceability is even mandatory, especially in the context of functional safety.
However, the quality of traceability is rarely questioned (except perhaps in an audit) and even more rarely is it actively used in development. This is a problem, because without quality, the added value of traceability fizzles out and lulls developers and decision-makers into a false sense of security.
Quality hardly considered in traceability
In a recently published paper Edouard Batot et. al. wanted to develop a metamodel for the Traceability create. The idea is not new, so the authors began by searching through relevant publications for traceability metamodels. The original aim was to identify a common core of all metamodels. However, the authors noticed that no metamodel addressed the quality of the traces.
Trustworthy traceability requires defining quality characteristics for the links.
Edouard Batot
However, the quality of the traceability has a direct influence on the use cases used for this. In addition, the quality degrades during development: every time someone changes, renames, moves or deletes an artifact, the traceability should actually be updated. Realistically, this is the exception, not the rule.
This problem has recently become more acute: Compliance-The requirements of the European Union's product development guidelines contain increasingly stringent traceability requirements with the aim of ensuring the functional safety of systems. At the same time, the number of traces in product development artifacts is increasing, driven by growing complexity.
Teams are introducing automation for traceability in order to meet compliance requirements. There are many approaches here, for example traceability based on naming conventions or through the use of AI. However, these approaches reduce confidence in the quality of traceability even further.
Quantify quality
To deal with this situation, the authors suggest tracking the quality of the traceability as a property of the traces. To do this, you have the quality for integral part of their metamodel made. For example, if we were to track the probability of the correctness of traces, then an impact analysis would contain a confidence interval.
That sounds like a lot of extra work. But as a first step, it would help to classify quality according to origin: A trace resulting from a review with humans is certainly of a higher quality than one created by an AI system.
Still a lot to do
Research here is still in its infancy. It will still be a while before it will be possible to systematically track the quality of traceability. But we will certainly have to do something here, at least for systems that place a high demand on functional safety.
Image source: Pixabay
