This is what I'm henceforth calling a (partially) Blindly Supervised Analysis, of Europeans and Caucasians versus the "Siberian Belt" peoples (excluding the East Asians).
The set-up was very simple. I set up the following poles on the populations presented (no other populations were included in the run).
1. Egyptians as the Fertile Crescent Pole. This was not blind, but ADMIXTURE did pick up the Fertile Crescent element blindly before. Egyptians have both much "Mesopotamian" and "Nile Core" elements, and I would find it very odd if they had any "Forager A" indeed (unlike the Druze who may have residual segments since they have so much mtDNA X), so they're a good choice. Naturally "Egyptian-ness" in this setting means only Fertile Crescent (Western Eurasian) affinity.
2. San-NB (Namibian Sans): blind pole - will not correspond to the actual component found, but instead used to "fish out" an unknown component. Obviously none of these populations are specially related to the San.
3. Mbuti Pygmy: another blind pole
4. !kung: blind pole
5. Hadza: blind pole
6. Yoruba: I included it because Mexicans have a small but significant chunk of West African ancestry. I don't expect much from Yoruba pole in Fertile Crescenters since there's Egyptian
1. San is about possible West "Siberian"/Amerindian/European Aboriginal components
2. Mbuti Pygmy didn't pick up much
3. !kung: I think this picked up specially "concentrated" "Nile Core" segments
4. Hadza is Eastern "Siberian" or Siberian proper and is prevalent in actual Siberians
5. Egyptian is really the Fertile Crescent Neolithic components
6. Yoruba is picking West African segments in Mexicans, and maybe also some "Nile Core" segments also found in Yorubas but maybe less frequently in Egyptians (not all the Egyptian, especially ancient, variability is represented in the individuals used as the pole)
Percentages are a bit different from other runs since all runs depend on the individuals actually present in the data. But there's a tight correlation. All the data I fed ADMIXTURE is the populations represented in the graph plus pole populations.
I did not remove admixed individuals, which should make this outcome less likely not more. I'm coming to the conclusion there is no way I can run ADMIXTURE with Europeans and Caucasians and Amerindians without some "Amerindian" appearing in both. With the Cypriots and Druze as FC pole it's more chunky but it's there. With Egyptians it's more even. In some runs Basques and Spaniards have much of it, in others less so. But Northern Europeans always have it as expected in a Neolithic (near) total replacement model, since they're the periphery of the expansion.
Chuvashes, Russians and Turks always come with "regular Siberian" admixture in all runs. I don't know how else to explain these results. Supervised, unsupervised, they're there. ADMIXTURE may not be that sensitive with small components. But I can't imagine why it would make "noise" consistently in the very same direction. "Noise" by definition is non discriminating. I think the Solutrean Hypothesis deserves greater consideration.
Amerindians almost certainly did cross Beringia. But just maybe, they were going the "wrong" way...