Estimating the number of faults remaining in software code documents inspected with iterative code reviews

Abstract

Code review is considered an efficient method for detecting faults in a software code document. The number of faults not detected by the review should be small. Current methods for estimating this number assume reviews with several inspectors, but there are many cases where it is practical to employ only two inspectors. Sufficiently accurate estimates may be obtained by two inspectors employing an iterative code review (ICR) process. This paper introduces a new estimator for the number of undetected faults in an ICR process, so the process may be stopped when a satisfactory result is estimated. This technique employs the Kantorowitz estimator for N-fold inspections, where the N teams are replaced by N reviews. The estimator was tested for three years in an industrial project, where it produced satisfactory results. More experiments are needed in order to fully evaluate the approach.