[T]o obtain a customer’s ethnicity estimate, we divide the customer’s genome into small windows. For each window we assign a single population to the DNA within that window inherited from each parent, one population for each parental haplotype. Each window gets a population assignment based on how well it matches genomes in the reference panel.

We do not know the exact haplotype boundaries, which differ between people, but we can achieve a good approximation by dividing the genome into 1,001 small windows. Each window covers one section of a single chromosome and is small enough (e.g., 3–10 centimorgans) that both the maternal and paternal haplotype, the DNA from Mom and the DNA from Dad, in a given window are likely to each come from a single, though not necessarily the same, population.