Each percentage is a share of the 4,046 markers we could read in your file (32% of 12,770), compared with people sampled today by the 1000 Genomes Project. The solid part of a bar is the share we are sure of. The faint part is how much higher it might be. The line is our best estimate. The range is a 95% interval from resampling your markers. If a population is missing from our reference set, your ancestry from it is counted under the nearest region we do have.

Each region and its range. East Asian 49.5% (likely 42–56%); Oceanian 42.3% (likely 38–46%); South Asian 8.2% (likely 2–16%).

South Asian is one region. These seven regional components share 96–99% of their allele frequencies, so how much South Asian ancestry you have is measured far more precisely than how it divides between the seven. Its total (8.2%) is firmer than the split inside it. The parts drawn inside it add to 0.0%, because some are too small to tell from zero and are not drawn.

East Asian is one region. How much of your ancestry is from this region is measured from the whole panel at once. How it divides between the areas inside it is a second, harder question, fitted separately — so the total is the firmer number of the two. Its total (49.5%) is firmer than the split inside it. The parts drawn inside it add to 38.9%, because some are too small to tell from zero and are not drawn.

Oceanian is one region. How much of your ancestry is from this region is measured from the whole panel at once. How it divides between the areas inside it is a second, harder question, fitted separately — so the total is the firmer number of the two. Its total (42.3%) is firmer than the split inside it. The parts drawn inside it add to 40.0%, because some are too small to tell from zero and are not drawn.

What “could not be placed” means. A region whose range reaches zero is not drawn on its own. The ancestry is still yours; this file does not have enough markers to say which region it belongs to. Here the regions that could hold it are African (up to 2%).

The map. Shading marks where each ancestry lives, at your own share of it. The outlines are geographic regions and the numbers are genetic, so an edge is approximate: ancestry shades into its neighbours. A faint region is one this file cannot tell from zero.

Paternal. Descent through the published haplogroup tree, entering a branch only where the marker defining it was genotyped and derived. 281 informative markers. Why it stops at K2b: the chip this file came from carries no marker that separates the branches below this point — it is not something missing from your file, and a test on a chip with denser Y coverage could go further. The full label K-M1221 changes as the tree is revised; the marker name does not.

An empty region on a map is one nobody has excavated, or excavated and not published.

Not found: we read the position and the variant is on neither copy. That is a measurement. Found, one copy or two copies: also a measurement. Not read: the chip does not type this position, or could not read it cleanly. Nobody looked, so it is not the same as not found. Could not be read: several variants share this position or the strand is unmeasured, so the row says what blocked it.

Most associations here were established in European or East Asian cohorts. Each row's tags say which population the association has been tested in. Catalogue v0.1.0. Nothing here is a prediction about you.

Rows report variant presence against published work. The catalogue (v0.1.0) is a small selection of the variants known in each gene: ClinVar records hundreds more, and roughly half of those are deletions, duplications or repeat expansions that no genotyping chip can detect at all. Metaboliser status for CYP2C19 or TPMT needs a full star-allele diplotype and an enzyme activity test, which this file cannot give. G6PD deficiency is diagnosed by an enzyme test, not a genotype.

Whether a position can be read depends on the vendor and chip version that produced the upload, and it is measured for each file. Frequencies shown are from gnomAD v4 and the 1000 Genomes Project. The evidence for an effect often comes from European studies, and each row's tags say where it has been tested.

Each file and every individual reference genome is projected into the first ten principal components of our panel. The chart is Mahalanobis distance in that space, so it scales each group by its own spread. The green band is where a typical member sits, the amber band is close but outside it, and a marker farther right is less like that group. Hover or tap a row for its distance. The closest areas come from the nearest three individual references, coarsened and left unranked.

On the same 289 people and 54 groups, this distance placed the correct group first 60.2% of the time against 35.3% for G25, and in the top three 74.0% against 63.0%. The G25 coordinates in this report are an export for other tools.

The axes come from the reference panel alone. Your file is projected onto them using the markers it shares with the panel, so it moves no point on the map.

Nearest is measured over more axes than the two drawn, to the middle of each group, so a group that looks close here can rank lower in the list.

The axes come from the ancient individuals alone. Your file is projected onto them using the sites it shares with them, so it moves no point on the map.

Nearest is measured over the first four axes, to the middle of each source. The plot shows two of them, so a source that looks close here can rank lower in the list.

On eleven test genomes the nearest source fell in the same region as the qpAdm fit. Inside a region, the order of the nearest sources does not track the qpAdm proportions.

Runs of homozygosity need no reference panel. The question is whether your two copies match, so this runs on 540,179 of your own markers, more than the 12,770 that reach the ancestry estimate. It is the one chapter in this report that our reference data does not limit. In total the stretches cover 76.4 cM of 3,545 cM.

Two different detectors were run over your file and they agree on 96% of the length. They fail in different ways, so where they agree the result is not an artefact of either one.

Lengths are in centimorgans, read off a recombination map, not in megabases. Near the middle of a chromosome a megabase can be a fifth of a centimorgan and near the end it can be two, so a generation count computed from physical distance is wrong by a factor that changes along every chromosome. Generations follow from genetic length as roughly 50 divided by the length in centimorgans, and a year figure uses 28 years to a generation.

Each stretch is shown as a range of generations. A 20 cM stretch averages about two and a half generations back, but over a very wide spread, and one stretch is a single draw from it. A range is what the evidence supports.

What this does not say. It says nothing about whether your parents are related. That question needs a comparison against the population you descend from, and we do not publish it. In a community that has married within itself for centuries the background is high with no close relatedness at all, so a fixed threshold would tell many people something untrue about their own family.

Heterozygosity is 24.8%: the share of read positions where your two copies differ. A consumer chip usually reads between 24% and 31%; well outside that range points to a file problem rather than to anything about you.

Your file is kept and used to build future reference panels, under a Creative Commons licence. We never sell or release an individual genome, and no genetic data is ever written to a log.

Accuracy is the share of held-out people whose largest region came out right: 120 people from 24 groups at the full panel of 12,770 markers, and again at 10,317 markers, the density a real 23andMe export reaches. The people tested are part of the reference set they are scored against, so their own data helped define the groups. Someone uploading a file gets no such help, so the real-world rate is a little lower. Five individuals per population. A perfect score on 120 people is consistent with a true rate above about 97%. The Indigenous American component is built from four samples, three of them majority European, so a correct call there is an easier test than it sounds. It measures which region comes out largest. Whether the printed range covers the true value is a separate measurement we have not run. 10 of the held-out set are not scored (African Caribbean and African American samples, which define no component). Measured 2026-09-09 against the 12,770-marker panel this report used. A rebuilt panel re-runs it.

A few dozen people per group is what makes our ranges as wide as they are. A narrower range would need more people. West Asia is fitted from Bedouin, Palestinian and Druze samples, and Oceania from pooled Papuan and Bougainville individuals. Central Asia is not yet a region of its own.