Vahaduo is the free, browser-based tool most people reach for the moment they receive a Global25 coordinate row. It is fast, it runs entirely on your own machine, and it does three jobs well: distance ranking, admixture fitting and PCA. It also has no guard rails, which is why the same tool produces both careful analyses and confident nonsense.
This tutorial walks through it as of 2026. If you do not yet have a G25 row, start with What are Global25 coordinates? and How to get your Global25 coordinates; nothing below works on a simulated row. A factual side-by-side of Vahaduo and our own tools, verified against the site, is at /compare/vahaduo.
What you need#
Two things, both plain text:
- A target — your own coordinate row, one line: a label followed by 25 comma-separated numbers.
- A source panel — the reference rows you want to compare against, one row per line in the same format.
Both must be in the same form, scaled or unscaled. Most published panels and most people's targets are scaled; if you paste a scaled target against an unscaled panel you will get a distance list that looks fine and means nothing. This is the mistake to check for first whenever results look strange — the full scaled-versus-unscaled story, including how to tell which form a mystery row is, has its own guide.
The layout#
Vahaduo's G25 views are a set of tabs over two text boxes. The Source box holds the reference rows; the Target box holds the rows you want analysed. Everything else — the Distance, Single, Multi and PCA tabs — reads from those two boxes. Nothing is uploaded; close the tab and it is gone.
Paste your source panel first, then your target, then choose a tab.
Distance: the closest reference rows#
The Distance tab sorts every source row by its Euclidean distance to the target across all 25 dimensions, nearest first. This is the most honest view in the tool, and the one to look at before any admixture fitting: it tells you what your row is actually near.
Read it with two cautions. First, distances are only comparable within one panel; a distance of 0.03 to a Bronze Age population and 0.03 to a modern one are not the same statement, because the panels differ in how their populations were averaged. Second, nearest is not descended from — the closest row is the average that happens to lie nearest in a 25-dimensional space, which is a similarity, not a genealogy.
Our free G25 distance calculator does the same computation against curated era panels with the dates stated, if you want a reference to check your Vahaduo panel against.
Single: one target, one mixture#
The Single tab fits the target as a weighted mixture of the source rows: it searches for the combination of source coordinates whose weighted average lands nearest your row, and reports each source's percentage plus a fit distance — how far the best mixture still sits from you.
This is where most misuse happens, so three rules:
- Fit distance falls as you add sources. Offer thirty rows and the fit will be excellent whatever they are, because the algorithm has more freedom. A low fit with many sources proves nothing; a low fit with three or four well-chosen, era-coherent sources is a finding.
- The sources decide the answer. Leave a relevant population out and its share is redistributed to whichever rows are nearest, silently. Include two near-identical populations and the fit will split ancestry between them arbitrarily.
- There is no p-value. Vahaduo cannot reject a model. If you offer sources that make no historical sense, you get percentages anyway. Formal rejection is what qpAdm provides — see qpAdm vs Global25.
A practical panel for a first pass is four to eight populations from one era, chosen because they plausibly precede the target and are distinct from one another. Then remove each one in turn and re-run: a source whose share collapses to zero when a neighbour is present was never earning its place.
Multi: several targets at once#
The Multi tab runs the same fit for every row in the Target box and tabulates the results. It is useful for comparing family members, or your row against a set of reference individuals, on the same panel. The same cautions apply to each row.
PCA: seeing the space#
The PCA tab plots the source and target rows on any two of the 25 dimensions. PC1 against PC2 shows the broadest structure; later pairs show finer regional structure. This is the view for sanity checks — a target that plots nowhere near the populations its admixture fit named is a sign the panel is wrong, or the target is unscaled against a scaled panel.
Our free G25 PCA viewer plots your row on curated era views beside individual ancient and modern samples, which is a useful complement: Vahaduo plots whatever you pasted, ours plots a fixed reference you did not choose.
Building a source panel#
The quality of everything above depends on the panel. Some practical rules:
- One era at a time. Mixing Neolithic and medieval sources in one panel invites the fit to use a later population as a proxy for an earlier one, which produces fluent nonsense.
- Prefer population averages built from several individuals. A "population" that is one low-coverage sample is a noisy point. If you build averages yourself, the free Average G25 tool does it in the browser.
- Keep sources distinct. Two rows within a tiny distance of each other are one source with two names.
- Write down the panel. A percentage without the panel it was fitted against cannot be compared with anything, including your own next run.
Our paid Global25 analysis applies exactly these rules through curated per-era calculators and prints every population the calculator was offered — used and unused — so the answer can be argued with. For the case where no curated panel fits a customer well, we hand-build one around their row; that is the Personalized G25 calculator.
Reading the fit distance#
A rough working scale for scaled coordinates: a fit distance below about 0.02 with a small, era-coherent panel is tight; 0.02–0.04 is ordinary; above that, either a relevant source is missing or the target is not well described by that era's populations at all. Treat these as bands, not thresholds — and remember the distance can always be driven down by adding sources, which is why the panel size belongs next to every fit you quote. The calibrated bands, the typicality baseline and the reasons lower stops being better are in the fit-distance guide; the panel-construction discipline behind them is in source selection and overfitting.
What Vahaduo cannot do#
It cannot tell you whether a model is admissible, cannot estimate uncertainty on a percentage, cannot tell a genuine row from a simulated one, and does not know which era its sources come from unless you do. Those are not flaws in the tool; they are the boundaries of coordinate fitting. The free G25 Authenticity Check covers the simulated-row problem, and How to read a Global25 report covers reading the numbers. The terms are in the glossary.
A five-minute routine#
- Confirm target and panel are both scaled.
- Distance tab first — note the nearest ten.
- Single tab with four to eight era-coherent sources; note the fit and the panel.
- Remove each source in turn; keep only those that survive.
- PCA tab on PC1/PC2 and PC3/PC4 as a sanity check.
Done properly, that is a defensible amateur analysis. Done with thirty sources and no era discipline, it is a random number with a decimal point.
