Why might a test fail even if the application works fine?