Dodecad is where all of this started. Dienekes Pontikos — a pseudonymous Greek blogger — launched the project in 2010, built calculators on the academic ADMIXTURE software, and released his DIY tooling publicly. Every other project on GEDmatch — Eurogenes, HarappaWorld, MDLP, puntDNAL — descends from that tooling. "Dodecad" is Greek for a group of twelve, and K12b, the project's most-used calculator, is the twelve-component model this post reads.
The twelve components#
Gedrosia · Caucasus · North European · Atlantic Med · Southwest Asian · Northwest African · East African · Sub-Saharan · South Asian · East Asian · Southeast Asian · Siberian.
As always in this family, the names mark where each statistical profile peaks in the training references, not places anyone's ancestors verifiably lived. Three of them deserve their own paragraphs, because K12b's readers spend most of their confusion on the same three.
Gedrosia — named for the ancient region spanning today's southern Pakistan and Iran — peaks in Baloch and Brahui references. It is nearly the same statistical object as HarappaWorld's "Baloch" component. In modern ancient-DNA language, the ancestry it tracks is Iranian-plateau-related — the broad ancestry stream that also spread with early farming toward South Asia. West Europeans carry a real slice of it (roughly ten points in some populations), which puzzled everyone in 2011; the ancient samples published since locate the source in steppe-related ancestry's own Caucasus-adjacent half, which K12b — built before those genomes existed — has no way to name.
Caucasus peaks where it says, and in 2011 it looked like a fog covering everything from Anatolia to Iran. Ancient DNA later split that fog into Caucasus hunter-gatherer and Iranian-related components; K12b's Caucasus blends them plus the Anatolian farmer legacy of the region. A large Caucasus score in a West Asian or southeast European profile is expected background, not evidence of ancestors in Georgia.
Atlantic Med peaks in Sardinians and Basques — the living populations richest in early European farmer ancestry. Read it as the farmer layer, with the same caution as everything here: the calculator predates the ancient genomes that proved what the component was tracking.
Reading a K12b breakdown#
The mechanics are the family's: an optimiser distributes your genome across twelve fixed profiles, always sums to 100%, attaches no error bars, and cannot reject a wrong panel. The K12b-specific advice:
- North European + Atlantic Med + Caucasus + Gedrosia is the West Eurasian quartet. Most European profiles are some blend of the four; the ratios are the informative part, and they track the deep layers (hunter-gatherer, farmer, steppe) better than any single number does.
- Southwest Asian is modal in Arabia; moderate values through the Mediterranean and Horn of Africa are ordinary. Northwest African at a few percent in Iberia is likewise expected.
- Trace East Asian / Siberian / Amerindian-adjacent slivers in West Eurasian profiles are noise or very old shared ancestry — not a recent ancestor. K12b has no mechanism to tell you which, and neither does any calculator.
- Compare within the calculator, not across. The same genome scores differently in K12b and HarappaWorld's similarly named components (a British kit runs ~44% North European in K12b but ~51% NE-Euro in HarappaWorld) because the components are anchored differently — why calculators disagree covers the mechanics.
The Oracle, and what replaced all this#
K12b pairs with the Oracle utilities — Dienekes wrote the original Oracle — which rank the project's reference populations by distance from your component profile. That ranking is the most interpretable thing a Dodecad run produces.
Fifteen years on, the exploration K12b pioneered runs against dated ancient reference panels instead of component constructs: era-scoped Global25 admixture calculators, a closest-populations distance ranking, and a PCA view of your coordinate among published ancient samples, free and in the browser. And when a component score graduates into a claim about descent, the tool with a test attached is a formal qpAdm model — the method that can say no, which no Dodecad descendant ever could.
Terms used here are defined in the glossary.
