You created a brand new HN account to shitpost about Rob Pike? I'm not even mad, that's amazing.

Does this bar serve anything other than everclear?

The idea is not bad but it sounds a bit extreme... Indeed choosing the right sorting algorithm is important and depends on one's problem to solve. But, 99.99% of the time, a general, all-round algorithm will do the job within 99% of the best-one's performance...

But when it fails, well, it will fail big... In my experience, some devs (me included :-)) make tests with a reduced dataset 'cos they can iterate faster and then go in production without testing on a real-size dataset. All of a sudden, the o(n²) edge case becomes, well, a problem :-)