Caro,
Value add is, for school within 'normal' ranges, a really good indication of what a school actually does.
HOWEVER, it is less reliable for supersuperselectives, because of 'ceiling effects' in the data. One of these effects works FOR a superselective school, the other works AGAINST, and as a combination they make value add for schools with exceptional intakes [I'm not talking 'normal' grammars, but those that take children from the top couple of %] unreliable.
The factor working 'for' these supersuperselectives is that the highest 'test' level at the end of KS2 was, until recently, a Level 5 - and this was used as the baseline for value add. For supersuperselectives, many of their intake will actually have been working - or have the potential to have been working - at L6 in Year 6, but until recently (certainly not for current GCSE / A level cohorts) this was rarely captured. So the grammar school 'gained' by children actually being assessed a level too low at the end of primary - they did not have to do anything to gain a Y7 full of L6 children, which would appear in the stats as a 1 level gain by the grammar school for those pupils.
The factor working against is that the highest GCSE mark is A. For a child in a school with a more normal distribution of ability, it is possible for that child to over-perform as well as under-perform - so given 100 pupils with the same level on entry, their levels at exit will have a spread, which crucially extends 'above expected' as well as 'below expected'. For a superselective only taking the top 1 or 2%, then expected grades on exit for the majority of children might well be A - none of those children can over-perform. Thus they get the 'downside' of children not getting their expected grades, but can't get the 'upside' of some exceeding expectations.
As i say, one positive and one negative - it would need a much better statistician than I am to work out which one has the more significant effect on the value add of a superselective in the Pates mould.