Larger classes should actually help. It is smaller classes that are affected most. I think I have mostly understood the way it works now, someone can correct me if it's not right but it seems to be the consensus online:
Last year a school had 20 people sitting A-Level maths. They got 4 As, 10 Bs, 3 Cs 2 Ds 1 U. This year the 20 people sitting in A-Level maths have received an internal ranking by the school. The top 4 (no matter what predicted grade they have been given) get an A, the next 10 get a B, the next 3 a C, the next 2 a D and the last one gets a U. The class this year might have people predicted A* but they can't have an A* because the class didn't get one last year. They might have 10 total predicted A/A* but only 4 to give out, 6 of them end up with Bs. They might have nobody worse than a C this year but because of last year the lower ranked people get Ds and U. It's not quite that simple but it's the gist. That's why people have been dropped multiple grades, it's based on previous performance of the school.
At GCSE then chances are that the school has more than 20 people, especially if it's a core subject. That means it is more likely to have a more even distribution of grades. I presume there will still be people downgraded but the chances of dropping multiple grades should be lower.
A low scoring cohort last year will badly affect this year or an exceptional cohort this year won't get the grades they deserve. Conversely, an exceptional cohort last year means this year's don't have to be as exceptional to get a decent grade.
Boromike you are right as far as it goes, I think. What would happen in a usual year is marks are normalised based on averages across the country, so if 10% get an A* last year across the whole country, then the top 10% define where A* ends and an A begins, and so on.
The problem with the algorithm is it is used on a school by school basis not an average across the country. This, as you say unfairly disadvantages a school that performed poorly in recent years, irrespective of how good the class of 2020 is.
I think the disparity could be larger in smaller class sizes just because they have a small representative sample, but that could also overestimate pupils achievements. The smaller the sample size the bigger the margin of error.
My belief is the algorithm was deliberately constructed to help private schools, and by default disadvantaged the government funded school system. The decision makers at OfQual may well have kids in private education and were biased, probably intentionally towards these institutions.
The main thing that is missed in all this, is what the government have done is illegal, there is no foundation to allow this to happen, other than an 80 seat majority, of course. GDPR specifically forbids this to happen, and does so clearly and unequivocally. The government do not have a legal leg to stand on.
This will result in many legal challenges and the government will, I think, do a U-Turn when they realize the scale of the issue they are facing.