Average G25 coordinates

Averaging Global25 rows means taking the arithmetic mean of each of the 25 dimensions independently: the first number of the result is the mean of every input's first number, and so on. That is how a population average is built from its member individuals, and it is why a population coordinate is a summary of people rather than a person. This tool does that in your browser and gives you a copy-ready row.

The reason to do it is that individuals scatter. One sample from a group can sit well away from where that group actually centres, so distances measured against a single individual carry that individual's noise as if it were the group's position. Averaging several members pulls the point toward the centre and makes the coordinate more stable to compare against.

The limit is that a mean is only meaningful when the inputs belong together. Average two genuinely distinct groups and you get a coordinate describing a population that never existed, sitting in the empty space between them — and it will still return confident-looking distances to everything. The arithmetic cannot tell you whether your inputs form one group; only knowing where the samples came from can. Averaging a handful of individuals from one site and period is sound; averaging by a shared label that spans centuries or regions usually is not.

Take four individuals from one archaeological site and period, pasted as four scaled rows. The averager returns a single copy-ready row whose every dimension is the mean of the four inputs' values — and that row now behaves differently from any member. Statistically, an average of N members carries roughly 1/√N of an individual's coordinate noise: four members halve the wobble, which is why a four-person site average makes a steadier distance reference than its steadiest member. It is also why our own reference panels enforce member floors — a 'population' of one is one person's noise wearing a population's name.

The checks worth running before using the result. Distance each member against the new average (the distance tool takes one paste): members of a genuine group typically sit within a few hundredths of their own centre, and one member sitting far out is either an outlier the label should not include or a data problem. And keep the form discipline absolute — averaging a scaled row with an unscaled one produces a coordinate for nothing that ever lived, with no warning from the arithmetic.

What the output is for: a steadier target for distances, a source candidate for a mixture panel, or the family-centre trick — average two parents and children should land near the midpoint, which doubles as a light sanity check on a household's kits. What it is never: a person, or evidence that the inputs belonged together in the first place.

The traditional way to build a G25 average is a spreadsheet: paste rows, AVERAGE each of 25 columns, reassemble the line by hand. It works, and every long-time hobbyist has done it; it is also where a specific family of silent errors lives — a column missed, a label row averaged in, precision truncated by the sheet's display settings, scaled and unscaled rows mixed in one range. The published community sheets are full of averages built exactly this way, most fine, some not, none marked.

This tool does the same arithmetic with the failure modes removed: full input precision, every dimension handled, nothing uploaded, and a copy-ready row out. The one thing it deliberately does not do is decide membership for you. Vahaduo-style workflows and this averager share that limit equally — no arithmetic can certify that your chosen rows form one population, and the era-scoped reference averages in our panels exist precisely because membership curation is analyst work, not spreadsheet work.