> quota system [...] You're optimizing for diversity and the OP is optimizing for quality of content.
I have never advocated for a quota system and I am not optimizing for diversity. Perhaps you should go argue with someone who holds those views, as I find this irritating and tiresome.
> There are 100 fibbwobblers. 10 are blue [...] 10 speakers [...]
And here you construct a ridiculous scenario to generate false intuitions. In particular, you are trying to generate the exact false intuition I pointed out with my first post: the notion that "there just aren't enough competent non-straight-white-dude speakers to fill out a conference list". It's horseshit. Peddle it elsewhere.
The only way to select speakers based on quality of content is to select speakers by content alone and hide all other criteria and characteristics of the speakers when selecting. Pretty much any other system introduces implicit and explicit biases, especially a quota system.
FWIW, yes, I'm claiming that your definition of better is different than that of the OP and that this difference is material. I may be wrong about this but based on your arguments in this thread and their arguments, if feels to me like you're both optimizing for different things. You're optimizing for diversity and the OP is optimizing for quality of content. The only way you get both is if the population from which you're drawing speakers at random has already been optimized for diversity.
There are 100 fibbwobblers. 10 are blue, 90 are green. Approximately 10% of fibbwobblers are great conference speakers (evenly distributed among all fibbwobblers regardless of color).
You want to hold a fibbwobbler conference with 10 speakers. To get only good speakers, you'd need to draw the 1 out of 10 blue fibbwobblers that are great speakers and the 9 that are green. If you have no quota system and end up with 10 green fibbwobblers speaking, approximately 1 of them is likely going to be bad speakers. If you decide to have a quota system where fibbwobblers of both colors are equally represented, 50/50, then you'll likely end up with the one great fibbwobbler speaking that is blue, and 4 blue ones that are not great speakers and 5 green ones that are great speakers. Under no quota, you get 1 bad speaker out of 10, under and equal representation quota system you get 4 bad speakers out of 10. Keep in mind that this is all based on the perfectly reasonable assumption that all fibbwobblers are equally capable of being a great speakers, regardless of their blueness or greenness.
Your worst case scenario in terms of quality of speakers gets comparatively worse as the spread between % of population that has identity X and the % of speakers that must be of identity X widens.
This is root of exception the OP was pointing out.