As I understood it:

1. Teachers submit both an estimated grade and a relative rank ordering for each student

2. Ofqual calculate a distribution based on the predicted grades

3. They compare it to grade distributions at that school from previous years

4. If need be, they shift the distribution up or down to be roughly in line with that school's prior results

5. They then use the rank ordering to assign students grades from that distribution (e.g. in a class with 20 students, each student is worth 5% of the distribution - so if it predicted 10% A* (highest grade), 15% A, etc., the top 2 students would get an A*, the next 3 an A and so on)