GEDmatch's admixture page is most people's first contact with component percentages: upload a raw DNA file, pick a project, pick a calculator, and get a breakdown in seconds. It is also one of the most confusing pages in genetic genealogy, because the projects are separate hobbyist efforts from 2011–2016, they use different components under similar names, and the site explains none of it.
This is the map. For what a calculator is in the first place — an optimiser that cannot say no — start with what is an admixture calculator; this post covers who built each GEDmatch project, which one suits which background, and what has aged.
Where the projects came from#
All the GEDmatch admixture projects descend from one lineage. Dienekes Pontikos (a pseudonymous Greek blogger) launched the Dodecad project in 2010 and released his DIY calculator tooling — built around the academic ADMIXTURE software — for others to use. Eurogenes (Davidski, the blogger behind the Eurogenes blog, who years later also created the Global25 system) adapted it with his own reference panels from 2011–2012. HarappaWorld (Zack Ajmal, 2011) adapted it for South Asian ancestry, MDLP (Vadim Verenich) for Eastern Europe, and puntDNAL for African and ancient-sample coverage. Each project's calculators were added to GEDmatch mainly between 2012 and 2016 and have been essentially frozen since.
That frozen date matters more than any other single fact here. The ancient-DNA revolution — Caucasus hunter-gatherers, the AADR reference panel, tens of thousands of dated genomes — mostly happened after these calculators were built. They are period instruments, still readable, but built before most of what we now know existed.
Which project for which background#
| Project | Built by | Best known for | Most-used calculators |
|---|---|---|---|
| Eurogenes | Davidski | European ancestry generally | K13, K15, K36 |
| Dodecad | Dienekes | West Eurasia, broad global sweeps | K12b, V3, World9 |
| HarappaWorld | Zack Ajmal | South Asian ancestry | HarappaWorld (one calculator, 16 components) |
| MDLP | Vadim Verenich | Eastern Europe, Balto-Slavic region | World-22, K23b |
| puntDNAL | puntDNAL project | African ancestry, early ancient-sample panels | K12 Ancient, K12 Modern |
Rules of thumb the community converged on, which our own reading supports: predominantly European ancestry → Eurogenes K13 or K15; South or Central Asian → HarappaWorld; East European or Uralic → MDLP as a second opinion; African → puntDNAL or Dodecad's Africa-focused calculators; mixed continental ancestry → run two projects and compare, because no single calculator is "the accurate one".
Reading any of them: three rules#
The components are constructs. "North Atlantic" (Eurogenes), "Atlantic Med" (Dodecad) and "NE-Euro" (HarappaWorld) are statistical profiles that peak in particular training samples, named by the person who built the calculator. Similar names across projects are not the same object: the same British genome scores roughly 44% North European in Dodecad K12b and roughly 51% NE-Euro in HarappaWorld — anchoring, not ancestry, moved those seven points.
The Oracle is where the meaning is. The percentage table is the least informative part of a GEDmatch run. The Oracle and Oracle-4 utilities compare your component profile against the project's reference populations and rank them by distance, which is the closest thing these tools have to context. The spreadsheet behind it shows what each component means in that project's own references — always worth a look before quoting a number.
Old calculators, old caveats. Some calculators their own authors have disowned — Davidski called the Eurogenes Jtest "only supposed to be a fun experiment… now horribly outdated" back in 2018. The pre-2015 ancient-sample calculators (Eurogenes Hunter-Gatherer vs Farmer among them) predate the discovery of Caucasus hunter-gatherer ancestry entirely, so their ancient components cannot represent it. When a calculator and a modern method disagree, the modern method is not automatically right — but a 2012 component set arguing with a 2020s ancient-DNA panel usually is losing for a reason.
What GEDmatch calculators cannot do#
The limits are the family's, not GEDmatch's: no standard errors, no test, no ability to reject a wrong panel, and trace percentages that are mostly noise. The full accuracy discussion covers what "accurate" can even mean for an optimiser. If a GEDmatch number is about to carry a real claim — an argument about whether a steppe-related source is required, whether a trace component is distinguishable from zero — that is a job for a method with a p-value: qpAdm, where a model can fail and the failure is informative.
Free modern alternatives#
If what you want from GEDmatch is the exploration rather than the nostalgia, the same exploration runs against current ancient reference panels in the browser: era-scoped Global25 admixture calculators, a closest-populations distance ranking across six eras from the Late Bronze Age to today, and a PCA view of where your coordinate falls among published ancient samples. Free GEDmatch alternatives compares the options, and how to get Global25 coordinates covers the one prerequisite the coordinate tools have.
Terms used here are defined in the glossary.
