Introduction

Be it our smartphones having to last through the day, automated industry plants, or autonomous robots and cars: We expect all these systems to make intelligent decisions to effectively and efficiently perform their tasks, regardless of encountered issues and changes in the environment.
There is a variety of techniques that we can draw on for implementing such behavior, including calculi like the situation calculus \cite{BRST00,GPRSF11} or dynamic planning concepts that allow us to react to changes in the environment \cite{CL17}.  Regardless of the technique, in principle we are searching for action sequences that allow us to achieve our current goals. In this context, we certainly not only have to check whether we correctly perceive and assess the environment, but in practice, we're also likely to suffer from unreliable actions. With our work as presented in this paper, we are focusing on the latter.
Reasoning about an issue's exact origin(s), e.g., with model-based diagnosis (MBD) \cite{rei87,kle87}, allows us to search for an ideal mitigation strategy. In practice, however, (a) there is seldom enough data to precisely isolate a problem's source(s) so that we end up with a set of candidates, and (b) the reasoning's complexity might prohibit us from making fast decisions. Moreover, in MBD we require in addition to observations of the system a system model that captures the behavior sufficiently to allow deriving diagnosis candidates. 
Complementing detailed and complete diagnosis concepts like MBD, in recent years spectrum-based fault localization (SFL) \cite{abreu09Barinel} has been gaining in attention. With SFL we evaluate execution data about which component was involved in this or that observed behavior. The result is a ranking how suspicious the individual components are to have caused the failing behaviors. Traditionally, SFL has been employed for software debugging, but, e.g., it was shown recently in \cite{PW18sfl} how to translate the idea for a static diagnosis of knowledge-bases used in automated reasoning. As will be discussed in the next section, there we observed which of the knowledge-base's rules were involved in the individual reasoning tasks and whether the tasks were successful. Because of the ease of implementing SFL and the smaller number of information required for diagnosis, SFL presents a representative of light-weighted diagnosis methodologies.