We hid the answer from the calculation, then counted what it got right
We took 81 families whose origin was known in advance, hid it from the calculation, and counted its hits and its misses. Everything is on this page, including where it gets things wrong.
Every figure is followed by the number of cases it was measured on.
Finding the right origin
How often the main region announced is the correct one.
These four numbers are read together. When a family stayed in one region, the calculation rarely misses. When its two sides come from different continents, two strong origins compete for first place and it can name the wrong one: that is where you should read the rest of your regions rather than the first one alone.
For scale: there are 17 major world regions to choose from. Answering at random would be right one time in 17.
How we measured
The principle is an exam with the answer sheet hidden. We gather families whose origin we already know, give the calculation nothing but birthplaces, and compare its answer with the known truth. It never sees the solution.
An exam like that has a trap: tune the method on the same cases long enough and it ends up knowing them. So we set 25 aside at the very beginning, and never looked at them during development. Those are the ones that give the honest measure, and that is the figure we put forward everywhere else on the site.
Compared with a DNA test
A few public figures have made their test results public. We ran our calculation on their birthplaces, without looking at their test, then compared.
These cases are few, and they are hard: all these people were born in the United States, where a birthplace says almost nothing about where a family came from.
A DNA test reads a genome. We read nothing of you: we start from places and replay the history of populations. On families born of the great deportations, a DNA test remains better, and no calculation will replace reading a genome. A consumer test itself carries 5 to 15% error.
The largest gaps corrected
Some lineages gain far more than the average. The four largest gaps between the old method and ours, across the 81 profiles of the bench. Each is a single profile, not an average: that is what the n beside it says.
Two profiles from the same bench went the other way and got worse: a Peruvian profile and a Somali one. They count in the overall figure at the top of this page.
What the calculation draws on
On 22,305 ancient genomes, the DNA of people who lived from prehistory to modern times, gathered in the AADR, the largest public database of its kind. And on 1,060 documented peoples, each placed back in its region and its century.
A limit we have not solved
Every figure on this page describes an average, over groups of families. None of them tells you how reliable your own result is. We do not know how to produce that measure yet.
The result still flags where it is weakest: a region with little reference ancient DNA carries a note, and the “~” sign marks the estimated percentages of ancient eras.
You know what the result is worth. Try it.
The preview is free and needs no credit card. Your first origins appear in a few minutes.
BeginA probabilistic lineage estimate, for cultural and recreational purposes, with no genetic or diagnostic value.