@Decorhate, of course they can't use the CAGs, that would be following Nicola Sturgeon. Much better to have stupid results! 
Re GCSEs, my initial thought was that 40% will change, so similar to A levels, but larger cohorts, so works better statistically. Then I realised that a whole load of schools always get 90%+ A/A* (selective state and private), so those will not be changed much at all. Which means that the 40% falls on the non-selective schools, and is more like 50% (guessing, I have not checked what proportion of schools are selective). Each subject is looked in isolation, so although on average people at non-selective schools might expect to see half their grades go down, there will be some who see none go down and others who see them all go down.
I think the momentum is now such that they will all end up with CAGs inflated by the changes the algorithm put up which can't be taken away once awarded.