Haplofield

People, sites and dates come from the Allen Ancient DNA Resource (AADR), release v66.p1, a curated collection of published ancient-genome metadata (CC0). Connections come from two places, kept apart. Measured ones come from ancIBD, published DNA segments that pairs of ancient people share by descent (CC BY 4.0). Reported ones are AADR’s “Family relations” column, the relatives each study reports.

The two are joined on the individual identifier both publish. Nothing is added at the join. A person with a site and a date but no shared segment stays a faint dot, and a shared segment whose two people have no published site is not drawn.

Reported relatives are not ancIBD, and Haplofield computes nothing from them. Each is what a study’s authors reported, by a method that varies from study to study. They stop at 3rd degree and carry no length of shared DNA, so the dial never uses them. They are also linked only within a study, so a missing pair means nothing.

Where both sources hold the same pair (402 pairs), the reported degree matches the length class of the ancIBD connection 87% of the time, and is at most one class off 100% of the time. No pair reported as 3rd degree appears in ancIBD, so that label, carried by 2,757 pairs, has not been checked.

The coastline under the space is Natural Earth, public domain. It gives you bearings. No dot is placed by it.

The published scene is content-addressed: build 4910f39d, published 2026-10-17. A new export is a new build with a new address, so a shared link always opens the scene it was shared from.

17,540 people carry a published site and are drawn. 4,103 of them appear in the pair file with at least one shared segment, and between them run 789,847 segments, each a stretch of DNA two people inherited from a common ancestor.

Dates run up to 185,000 years before present, and the median person lived 2,500 years ago. Each date is the published estimate with its published uncertainty, on average ±93 years, drawn as a bar along the time axis.

Sex, where the source records it, and the coarse length class of each connection are carried as published. The scene declares its own counts and ranges, and the viewer draws what the scene declares.

Build 4910f39d carries 8,300 reported pairs whose two people are both in the space, kept apart from the segments and off the dial. 7,898 of them are pairs ancIBD does not hold, and they give a first connection to 1,300 of the 4,866 faint people.

Another 5,491 of those pairs name at least one of 1,514 people ancIBD never tested, because too little of their genome was recovered to look for shared DNA. They are drawn as open rings. An open ring means nobody has looked. A faint filled dot means somebody looked and found nothing.

Every dot sits at three published coordinates: the longitude and latitude of the site where the person was found, and the date they lived. Nothing is projected, embedded, averaged or imputed. A person without a published site is not drawn at all.

The whole space rests on that. Position is archaeology: where a body was found, when it was buried. The dial is genetics: how much DNA two people share. The two are independent, so turning the dial recolours the space but never moves a dot.

Longitude and latitude are drawn on an equirectangular floor at one scale, so the coastline keeps its shape. Time runs up the vertical axis by rank, so the many recent people and the few very old ones both have room. The axis labels state that scale.

The dial is a threshold: the minimum amount of shared DNA, measured in centimorgans (cM), for two people to count as connected. Closer relatives share longer stretches, so a higher setting keeps only the closest links.

Where it bites is a property of the data. Below about 41.5 cM nearly every connected person sits in one undifferentiated group. Between 41.5 and 87 cM that group collapses into hundreds of separate ones. Above it, the groups are people buried in the same place within a century or two of each other: households, cemeteries, people who knew each other.

The still figures on the home page and the landing are drawn at 60 cM, inside that band, with one person selected and their connections drawn down to 20 cM. That is the view you get by pointing at anyone in the space. Every dial setting is reachable there, and the address bar records it, so a view can be shared exactly.

At any setting, a group is a connected component: the people you can reach from one another by following connections at or above the threshold. Each group takes a stable color. The largest group is drawn gray, so the many small ones read against it, and people with no connection at that setting stay faint but present.

A group is a statement about shared segments and nothing more. It is not a population, a culture, an ethnicity or a lineage, and its color carries no meaning beyond telling it apart from its neighbors.

Families also files people by the relatives their studies report. A reported pair stays together at every dial setting, which places 2,814 people the shared DNA alone could not, 6,591 in all. At 120 cM, 781 of its 989 rings rest on a report, and the key marks them. The dial and the groups it counts come from ancIBD alone.

Cite Haplofield with its build and publication date. Together they pin the exact scene you saw:

Grove Lab. Haplofield, build 4910f39d, published 2026-10-17. https://grove-lab.org/en/haplofield (accessed 2026-10-17).

Haplofield re-plots published data, so a claim about a person, a date, a shared segment or a reported relative should also cite where it comes from: the Allen Ancient DNA Resource (AADR) with its release version (v66.p1), the original publication for each individual you discuss (which is also where each reported relative was published), and ancIBD for the segments. Both datasets and their licenses are listed under Sources and licenses.

For a specific view, link the view itself: explore links (/haplofield/explore?…) carry the view and the dial setting, so the link you share opens where you were.

The build keeps a citation exact: a build is addressed by its content, so one build always holds the same people, dates and segments. Only the current build is online. When a new scene is published, links open on it, and the build in your citation still names the one you saw.

The space is only as complete as the record it draws:

• Coverage follows where archaeologists have dug and labs have sequenced. A dense region is one that has been studied, and an empty one may simply not have been.
• Dates are the published estimates, with an average uncertainty of ±93 years and much more for some individuals. A bar is drawn for each, but two people whose bars overlap are not thereby contemporaries.
• A threshold is a choice. The band where the space reorganizes is measured, but no single setting is the correct one.
• A group is a set of shared segments. It says nothing about who a person was, what language they spoke or what they called themselves.
• Reported relatives are only as complete as the studies that report them. They stop at 3rd degree, each study links only its own people, and the studies used different methods. A person with no reported relative may simply have been published without one.
• Haplofield makes no new biological claim. It re-plots published data under a published method, so anything it appears to show that is not in the sources is an artifact of the view.
• One published scene at a time. Build 4910f39d is the current one, and earlier builds are not kept online.