Is this all about about handling unanticipated changes or about foreseeing what needs handling?

I am not convinced that the challenge for self-adaptive software is how to handle unexpected changes. I believe that the real challenge is how to characterise the uncertainty that the self-adaptive software has to handle. Once a proper characterisation of the uncertainty is in place, I think handling unanticipated changes and situations becomes a lot simpler. What is difficult is to know what range of uncertainty that self-adaptive software should be designed for. To use an autonomous driving example: should we expect people to jump in front of the vehicle or can we ignore that part of the problem because we are driving in a freeway and there are supposed to be no people around? The ultimate challenge then becomes providing guarantees and legally valid contracts and understanding where the responsibility lays in case of errors.