OK, I think I finally understand.

What was said was:

E.g. I've often heard it claimed that the difference in average pay between women and men is mostly attributable to differences in ambition and time voluntarily spent at home with children. I haven't looked at the matter enough to know if this is true. But if it is, then denying any population-level differences between men and women seems harmful, because it implies that something that actually has an innocuous explanation is because of discrimination.

One common explanation of harm and utilities is that the "real" or important utility function held by a human is that implied by the humans actions. If a human chooses A over B, that means to the human A has a higher value than B to the human. This runs us into problems, for example when humans choose B over C and C over A, but there is no agreed upon way to discuss the relationship of humans to utility functions. We just don't know how to extract the human and cut the nonsense without cutting the human! This is despite extensively discussing extrapolate volition. One way to get people to actually choose consistently among A, B, and C is to teach them about this paradox, but let's just say for our purposes here that it's clearly not out of line to discuss people's "true" preferences being something other than what they choose.

Vaniver: Ok: let's suppose he intended the primary definition of innocuous, "not harmful." If a choice is made voluntarily, then by the assumption of revealed preferences it is the least 'harmful.' If we forced women to choose with the same distribution that men do, then on net women would be worse off- i.e. harmed by our force.

Clarica: I think that calling the choice to spend more or less time doing financially unrecompensed work in the home an innocuous gender difference, is careless. The harms of the various choices have not been evaluated that well.

One issue is that language is flexible, and it is common to see "innocuous explanation" as a way of discussing the motives of a person causing the things the explanation explains, rather than according to the usual adjective-noun relationship where the adjective modifies the noun.

For example: a video teaching "how to fold a shirt" with the audio 50 decibels is a harmless explanation. The same video with the audio at 125 decibels is a harmful explanation.

No one argues that the explanation itself would have only good consequences, the discussion is instead what sort of harmlessness is meant instead. Whether the author's intent is clearly that, if it is discovered that women's actions alone cause the statistical difference, i) employers are doing no harm in the hypothetical case, or ii) if a similarly plausible interpretation is that no one is suffering harm, for had they chosen as men, there would be no disparity.

Context points to the first explanation as the best contrast with "discrimination", what employers are allegedly doing, and what hypothetical evidence would clear them of, but it's easy to see why someone intending the second point might have used the same words.

The sentence might be rewritten: "But if it is, then denying any population-level differences between men and women seems harmful, because it implies that something fully explained by innocuous behavior is because of discrimination."

The principle of charity protects us in similar cases where we happen to only see one interpretation and it is the wrong one.