Why does the same person get different G25 numbers from different calculators?
If you have ever pasted your Global25 coordinates into two different tools and gotten two different sets of numbers back for the same file, you are not imagining things and you did not do anything wrong. It is one of the most common questions we get, and it almost always comes down to one of three things: scaling, the reference panel behind the calculator, or how your raw file was converted to G25 in the first place.
Scaled vs unscaled: the most common mismatch
Global25 coordinates come in two flavours. The "unscaled" version keeps the coordinates in their natural, very small range, usually well under 0.02 in absolute value. The "scaled" version multiplies everything up so the numbers are easier to work with and compare visually, typically landing somewhere between -0.3 and 0.3.
These are not two different results, they are the same underlying data at a different scale. But if you paste an unscaled set into a tool expecting scaled coordinates (or the other way around), every distance calculation the tool runs will be wrong, sometimes wildly so. Before comparing two sets, always check they are both the same type.
Raw file vs official G25: what changes when you convert
If one of your sets came from uploading your raw DNA file directly and the other came from a calculator that used Davidski's official Global25 pipeline, small differences are expected even when everything is done correctly. Converting a raw genotype file to G25 involves imputation and projection onto a fixed set of reference axes, and different pipelines make slightly different choices at that step. The two results should be close, usually well within a few thousandths on most axes, but rarely identical to the last decimal.
A bigger gap, especially concentrated in one or two specific axes rather than spread evenly, is more often a sign of a genuinely different processing choice (a different chip, a different imputation reference, or a bug) than of "phasing noise."
How to actually compare two sets instead of eyeballing them
Manually scanning 25 columns of decimals to spot where two coordinate sets disagree is slow and error-prone. We built a free tool for exactly this: paste in two to six named G25 sets for the same person and it will show you, side by side, which modern populations each set is closest to and, more usefully, which of the 25 axes actually carry the disagreement. In our experience the first two or three axes (which capture the big north-south and east-west European gradients) almost always agree closely between sources. The real divergence is usually hidden much further down the list, on axes most people never look at.
You can try it here: Compare G25 Coordinates. It is free, nothing is stored, and it works with as many named sets as you want to paste in.
When the numbers really do disagree, and what to do about it
If the comparison shows a consistent gap across most axes rather than one or two outliers, the most likely explanation is that one of the sets was generated from a lower-quality or older chip, or from a heavily imputed low-coverage file. In that case, trust the set built from the more complete raw file, and treat the other as a rough approximation rather than a contradiction.
If you would rather have a full ancestry breakdown built directly from your raw file, with the modelling and haplogroup detection handled for you, our reports accept either a raw DNA file, a pasted G25 set, or both together, the raw file is always used for haplogroup detection even when you provide your own G25 coordinates for the modelling.