Interesting replies here.
So, Marcia and jcoronella, you're both implicitly saying that the top 1000 is probably "sort of right", ie. that the top 1% is most likely found within that set - ie. not outside it?
Or perhaps it didn't cross your mind that the 1000 was just a sample of the students of that discipline from all over the world? Sometimes there are millions, sometimes a few - but still it sounds as if you've got a feeling of some kind that the top 1000 (sorted on grades alone) is, well, sort of "not all that wrong"?
I'll buy that assumption for now. There'll be misses, of course, especially related to broad disciplines (eg. "The Sciences"), but of course, the more specific your task/query is (eg. "Quantum mechanics and String Theory"), the easier it will be to identify the top 1000 properly, as the total pool of students will get smaller.
Now, that's interesting. I don't know if it was the wording of my question that did it, but... both of you intuitively would put the top 1000 through an extra test?!
Ie. you wouldn't "call a friend" or something. In stead, you'd assume, instinctively, that the answer was found within that (sub-)dataset already, and that it (by itself) could be used to get a more accurate ranking.
Perhaps it's even as simple as asking (some or all) of these 1000 students who they would recommend themselves?
Now, that's a simple solution to me ... I'll buy it.
Also, i've got this odd feeling, that 1) who you ask, 2) how you phrase the question, and 3) what you do with the answer, will determine if you should call it Hilltop, or LocalRank, or whatever...