ASTRAR provides a framework for testing Web Services (WS) using the group testing technique. This paper extends the basic two-phase testing process and introduces the windowing mechanism to further improve testing efficiency. Rather than testing a large number of WS simultaneously, WS are divided into subsets called windows and testing is exercised window by window. Testing results are analyzed for different strategies such as using all of the historical data, using the most recent windows, and using the current window only. Based on the results, test cases are ranked according to their potency to detect faults; and oracles and the confidence level of each oracle are established for individual test cases at run-time. In addition, different strategies are proposed to determine the optimal window size at run-time. By incorporating the windowing mechanism, the two-phase training and volume testing process becomes a continuous learning process and the basic group testing process becomes more adaptive to dynamically changing environment.