I think that's part of the point. i.e. the usual line about wanting to avoid false positives (bad engineers putting on an act) even if it means sacrificing many false negatives (good engineers who suck at interviewing, or even just simply wasn't having a good day on the day of the interview).

Ironically the modern technical whiteboarding interview often does require the candidate to effectively act and put on a show,

I knocked together a little simulation.

Suppose programming ability is normally distributed with a mean of 100 and and SD of 15 (values cribbed from IQ). A team starts with 10 people drawn from that population.

More applicants are drawn from that population: if an applicant's ability is above the median of the current team, they are hired, replacing a randomly-selected member of the team; otherwise, we pass.

At the first timestep, half the applicants get hired right away. It takes a few hundred total failed interviews to make the 25th hire, and a few hundred thousand(!) for the 100th hire. These have very long tails too: the longest run (of 1000) took 800M total interviews to hire 100 people.

There's no way this rule is actually being applied in practice.

raising the bar would also mean that their pool of people who are merely good enough gets very small very fast and Amazon just doesn't have the reputation to attract the absolute best