Small, unexpected percentages — like 2–5% of a region you have no known connection to — are often what we call "random noise." This happens because our algorithm sees genetic similarities between certain populations and assigns a small percentage where there may not be actual ancestry.
For larger discrepancies (like seeing 70% of a region that doesn't match your family history), it's usually because we don't yet have the right labels or enough reference samples for your specific background. For example, if you're North Indian, your results may show as "South Indian" because we don't have a separate North Indian label yet. We're adding more specific labels and samples in our next update.