Imagine you have a database table with a trillion examples describing people, and tens of thousands of columns expressing various qualities about the person row such as but not limited to strength, speed, looks, etc.
Also assume the examples completely explore every single range of every column, said another way, they are evenly and uniformly distributed so that you can find anything you want, but the cells were defined more or less randomly, and some rows are mutually exclusive to other rows, such as $strength $intelligence, and $speed, you can only get one, but not the other one, for obvious reasons, the strongest person isn't the fastest person.
I give you the option to make as many rank_order() methods as you like to separate which one is best. You give me your choices and I short-list this way:
select * from person order by strength desc, intelligence desc, speed desc, looks desc, etc limit 10;
You get ten rows but you're unhappy because your first criteria contain the strongest people to have ever lived and therefore the next two properties are below the median.
So again you re-order your asserts:
select * from person order by looks desc, speed desc, strength desc, etc limit 10;
Again you're even more unhappy, you're staring in the face of the most hypnotizing intoxicating fairy to have ever lived, but now it has below median on strength, speed, int, etc.
When ever you rank order a population by a thing, everything else that thing is remotely inverse correlated with, such as but not limited to intelligence and speed, goes below the median, and if you try to get clever and rank order after a sum, you're still failed because you have not correctly expressed the variable skew in the decay function for when strength starts decaying from intelligence. Which is opposite decay, from 100 to 115 than it is from increasing IQ 75 to IQ 95.