I find it interesting how much the memo is misinterpreted even by folks who have read it multiple times and have been pretty engaged in discussions about it. Example:
"He claimed that Google’s diversity efforts represent a lowering of the bar. Google has stated many times that its efforts involve focusing more resources on searching for candidates in minority groups rather than lowering the bar for these groups."
This is what he actually wrote:
"Hiring practices which can effectively lower the bar for “diversity” candidates by
decreasing the false negative rate"
I'll admit this is a surprisingly dense sentence. And the phrase "lowering the bar" is controversial for an obvious reason, so it was a really bad decision by Damore to combine such a phrase with such a dense line of reasoning. It's just a recipe for disaster.
Some things to note:
* The use of "effectively" here is critical and means that the bar is not directly lowered in interview or technical standards. Rather, it's lowered as a (potentially complex) side effect of other more indirect policies.
* That indirect policy is to decrease the false negative rate. This doesn't mean standards are actually lowered for minority candidates; rather, it means extra time and money are invested to make sure that a negative result is actually a negative result. This can cause the standards to be effectively lowered, because it means that extra time and money are not invested in non-minority candidates, who therefore must overcome a higher false negative rate -- and the easiest way to do that is to be so good that none of your interviewers might feel ambiguous about your abilities.
To state that second bullet point differently and in a way that totally avoids politics, if you imagine that the interview is producing a noisy estimate of your technical ability, and the noise is predominantly negative/subtractive, then the expected score of an accepted candidate is higher with the noise than without. But if you offer a second interview to candidates and let a candidate's final estimate be the greater of the two samples, then the negative noise is reduced (though not eliminated), which will lower the expectation (but NOT below the expected value you'd get by eliminating the noise entirely).
With this analysis in place, I think it's more helpful to rephrase it as effectively raising the bar for non-minority candidates rather than lowering it for minority candidates, though ideally I'd prefer to avoid any phrases about bars being lowered or raised in general.