What you say can only be partially correct. If I pick any random Habashi/Afar/Oromo population and Somali and include Yemeni also in the source and target myself, a Somali with some recent Yemeni ancestry, the calculator will automatically shift toward the Ethiopian populations and regard the Somali ancestry as lesser. We can be wrong in assuming the calculator is correct in discerning all backgrounds in equal regards.
One fun illustration is if I overfit and only use Ethiopian populations without any Arab or Somali, that will give me a closer fit than Somali samples, adding all Yemeni/Saudi samples. That is, Vahaduo will deem it closer to my ancestry than anything regarding Somali+Yemen. Granted, the third factor at play is, namely, that my ¼ Yemeni ancestry carried some Ethiopian-like (seemingly similar to Mota, not Cushitic), which undoubtedly is at play of skewing things further.
But what's interesting in particular is that it only removes the South-Arabian in Somalis. Check for yourself. The mota ancestry mostly stays, barely any of the "Natufian" or "Dinka" is touched and yet the Yemeni completely disappears. And Horner mediated admixture just makes more sense in terms of uniparentals and history in my opinion.