Some relevant criticism, some not

I hadn't thought about this before, but Seamus has some points.

First, as modernites, we are used to see the children fully dressed, and we are used to shortcut any natural way to tell them apart by appealing to their dress, which is cultural. So we are (intentionally by the culture, imho) made ignorant of natural boy/girl differences and therefore we fail to look for them and recognize them. A neutral test would have to use either "primitive" peoples (or "dirt poor" populations) who are used to see children fully naked or naked except their genital region (whether modern underwear and shorts or traditional loinclothes), and who therefore are used to notice body clues that we fully-dressed modernites are not. Essentially, your sample is not relevant to humankind. It's only relevant, at best, to fully-dressed modernites.

(And I say this as someone who has intentionally studied the body differences between boys and girls and knows how much he is losing in perception because they're all dressed)

Second, you did post a relatively large number of intentionally ambiguous or androgynous children. A truly random sample of children would always have many fewer "androgynous" children than you posted. So, your test proved that ambiguous children do exist. Also that neither a GL nor a BL nor a teleio population are significantly more apt at telling them apart. That alone is a good (and replicable) finding. But it does seed doubt into your 50% rate. The rate is artificially low by your intentionally "hard" sample used. Life has a much higher success rate - I know nobody that fails 50% of the time in real life, seriously. Do you?

Third, naked chests might be not publishable on GC, but I do believe that, even if it is very vestigial compared to what it should, and unconscious to a point, we are better when we see the whole person than the face only. I have a feeling if your test had not been a face test but a full body test, you'd get a better answer rate. Now of course, there is the problem that clothes send strong cultural messages, so it is extremely hard to isolate that factor; but I am willing to bet that if we dressed, idk, a hundred children identically (whether all as boys, all as girls, or all with a reasonably gender-neutral attire), gender distributed 50/50, and with a large proportion of ambiguous looking ones as in your original test - you would find out that face alone got a lower success rate than whole body. In any case I think it would be a more interesting test. And moreover, I think it would reflect the tested subject's real exposure to boy/girl differences that has learned, consciously or subconsciously, to tell them apart by body.

And finally, you two (or you three) are bound to be measuring different things. The boy/girl differences in body are given by gender, and they are derived from bone structure, therefore they exist, as I said, even before birth. Any anthropologist and a good forensic can tell apart a dead boy and a dead girl just that way. But face differences depend much less on gender and much more on race. (And the same anthropologist will have a much harder time telling you the gender of the child from a skull alone - but will be able to tell you their race) The chests of White, Black and Asian girls have more in common with each other girl than with White, Black and Asian boys. But the faces of White, Black and Asian girls have more in common with White, Black and Asian boys than with each other. Bodies are more similar within gender across race; faces are more similar within race across gender. It would be still more interesting, then, to make your test cross-racial. I did notice too many Whites the first time. :p More scientifically important, it would be a way to test whether it partly depends on familiarity (as I am openly speculating). Assuming most GCers are more used to White people, it should follow that success rates are higher in the White subset of the test.