Page 6 of the notes (Chapter 17 Modelling) says when listing potential operational issues: A range of methods of implementation should be available to facilitate testing, parameterisation and focus of results. What does a 'method of implementation' mean in the context of an actuarial model?
Another way of expressing this point would be to say that a good model can be run in a number of different modes. Maybe there's a "debug" mode where the model outputs an abundance of intermediate calculations to facilitate testing which it might not do in the regular "production" mode. Maybe the model has the ability to be run using a test pack of data instead of the full data set. From an IT point of view, maybe the model can be run on a desktop PC rather than requiring the full might of external cloud computing. What's being avoided is a model which operates as a black box which can only be run in a single way.
