Issues regarding black box, large systems verification are explored. It begins by collecting data from several testing teams. An integrated database containing test, fault, repair, and source file information is generated. Intuitive effectiveness measures are generated using conventional black box testing results analysis methods. Conventional analysts methods indicate that the testing was effective in the sense that as more tests were run, more faults were found. Average behavior and individual data points are analyzed. The data is categorized and average behavior shows a very wide variation in number of tests run and in pass rates (pass rates ranged from 71 percent to 98 percent). The 'white box' data contained in the integrated database is studied in detail. Conservative measures of effectiveness are discussed. Testing efficiency (ratio of repairs to number of tests) is measured at 3 percent, fault record effectiveness (ratio of repairs to fault records) is measured at 55 percent, and test script redundancy (ratio of number of failed tests to minimum number of tests needed to find the faults) ranges from 4.2 to 15.8. Error prone source files and subsystems are identified. A correlational mapping of test functional area to product subsystem is completed. A new adaptive testing process based on real-time generation of the integrated database is proposed.
