[jkic47]

I don't know that I agree with your first line (in which case all hiring managers would end up with essentially indistinguishable outcomes), but I want to address the next 2 paragraphs.

"Best" being a subset of "good" is not some obscure fallacy; but well recognized by HR departments. For the purposes of most jobs one hires for, they are well aware that Persons A and B are likely fungible quantities (and, also, why the impersonal word "resources" is in their very job function). We hire for demonstrated work ethic, ambition, the ability to deal with adversity, and work track record, rather than skin tone, genitalia, sexuality, handicaps, etc. Foreseeably, we end up matching (more or less) the demographics of the population we source from.

Being fallible humans, we absolutely do have blind spots in our hiring process; something we fight hard to correct for. I concede it is entirely possible that some experience a diverse person has had might some day become useful, but I am skeptical that similar experiences don't exist in most candidate slates. I am obviously overstating this for effect, but we have never done a lessons learned session and said "damn, if only we had hired a black lesbian quadriplegic, instead of John Smith, we would have succeeded at our task".

We hire those we perceive to be the best match from our slate of candidates. Some of these people are women, some are old, some are variously pigmented, but all of them can do the job they were hired for.

[jltsiren]

Most people end up being employed. Therefore most hires are average. A lot more people believe they are hiring above average candidates than actually do so. Good employees are rare, for any meaningful sense of the word.

There is also another common fallacy: you measure what can be measured easily and then use the data to make conclusions about what you would have wanted to measure. For example, you measure the performance of an employee and use that to make conclusions about the success of the hiring decision. But to actually determine that, you would have to know how well other candidates would have performed in the role, and how the presence of the chosen candidate (would have) affected the performance of everyone else. Among many other things.

That's also a big part of the reproducibility crisis in science. It's impossible to make justified conclusions from data alone. You can rarely measure what you actually wanted to measure, and you never know if you took every relevant factor into account. In order to make conclusions, you have to assume a model of the system you are measuring. Then your conclusions depend on the assumption that the model you have is a useful description of the system.