In the few minutes we have left before adjourning this session, I'd like to review briefly various points that have been made this morning. The development of a failure predictive technology evidently involves a proper combination of several ingredients. These include a quantitative measurement capability, a proper description of the mechanisms of failure, a way to combine these elements into a probabilistic failure predictor that takes account of the variance in both the above quantities, a way to include a priori data that may exist which will help to sharpen the prediction, and finally, a way so that management choices (which govern the cost-risk tradeoffs) can be injected into the predictions. This morning we have heard a number of talks relating to one of the above items, i.e., the mechanisms of failure in metallic materials with an emphasis upon fatigue processes including fatigue initiation, crack growth mechanisms, and complications that are introduced by consideration of the fully plast.ic problem.