An experimental study of fault detection in user requirements documents

This paper describes a software engineering experiment designed to confirm results from an earlier project which measured fault detection rates in user requirements documents (URD). The experiment described in this paper involves the creation of a standardized URD with a known number of injected faults of specific type. Nine independent inspection teams were given this URD with instructions to locate as many faults as possible using the N-fold requirements inspection technique developed by the authors. Results obtained from this experiment confirm earlier conclusions about the low rate of fault detection in requirements documents using formal inspections and the advantages to be gained using the N-fold inspection method. The experiment also provides new results concerning variability in inspection team performance and the relative difficulty of locating different classes of URD faults.