Discovery and repair of software anomalies
The average time required to discover and repair anomalies which appear randomly in the end-to-end tests performed after program completion is evaluated. A model which explains anomaly discovery and repair phenomena when applied to variations in work load and multiple-stage testing is presented. The theory shows that estimates of anomaly levels and team capability can be predicted after a significant fraction of the anomalies have been found, and indicates procedures for applying these figures to schedule estimation and work-load assignment. Of particular interest is the demonstration that end-to-end testing of programs in other than the operational environment is not generally cost effective.