Kurin’s Jefferson Preprint

22 September 2026

Kurin’s preprint is now posted online--“The DNA legacy of Thomas Jefferson” (bioRxiv, 21 September 2026)--attached. This is an initial assessment, drawing on the study’s own tables, a genealogist’s reading of the pedigrees, and the first outside technical review of the paper.

The standard this is measured against.

For a paternity claim with no documentary paper trail, the discipline genealogists apply is clear and convincing evidence of the blood line--which in this case could be met only by obtaining Thomas Jefferson’s own DNA and comparing it directly to the DNA of the Hemings descendants. The study set out to answer exactly that, and recovering a usable genome attributed to Jefferson is a genuine achievement. The question here is whether the proof they build on it meets that standard. In my view it does not yet, for the reasons below.

The strongest result depends on a new, unvalidated method.

The paper’s most powerful claim--the “less than one in a thousand” figure--depends chiefly on the Madison line, and specifically on long stretches of shared DNA between Madison’s descendants and the genome attributed to Thomas Jefferson. Colleagues at the Max Planck Institute, whose ancIBD method is the field standard for detecting shared DNA in ancient, low-coverage genomes, have made an initial review and report that these long segments are called only by the team’s own new pipeline and are not reproduced by the established method. To find shared DNA in so degraded a genome, Green’s team built a pipeline of their own that skips imputation--the standard, well-tested step that statistically fills the gaps in a lowcoverage genome, and the step the established method uses. If the long Madison segments do not survive that established method, the result the headline depends on is not confirmed. This is why the recovered genome and the underlying data need to be released, so the calls can be checked independently.

For Eston, the numbers are far weaker than the headline.

Their own likelihood ratio for Thomas against his brother Randolph is only about 5 to 7 (Table 1)--a posterior of about 83 to 86% for Thomas and 12 to 16% for Randolph. The “less than one in a thousand” figure is the joint probability that Thomas fathered at least one of Eston and Madison; the chance he fathered both, on neutral priors, is only about 82 to 86% (Table 2). The strength is in the Madison line, not Eston--and that Madison strength is exactly what the outside review now questions.

Their framing outruns their own numbers.

The introduction closes with “strong evidence that Thomas Jefferson himself, not a close relative, fathered multiple children with Sally Hemings”. Yet the results section says the father “must have been a close relative of Thomas Jefferson if not Thomas Jefferson himself”, and that for Eston the Randolph scenario “retains a smaller but considerable proportion of the posterior probability”. The confident exclusion of a close relative up front is not what their own model reports.

Thomas and Randolph are separated by simulation, not by DNA.

The study models Thomas, Randolph, and a son of Randolph, and includes 2 living Randolph-line descendants to calibrate, but there is no genome for Randolph himself. The two are told apart only by the amount of shared DNA the pedigree simulation predicts, not by a direct comparison to Randolph’s own DNA--and a full brother shares roughly half of Thomas’s genome, so the modelled gap is small.

These were not large, outbred families.

The analysis assumes that shared DNA between two people comes from the single relationship being tested; here that assumption is genuinely at risk. Thomas Jefferson’s wife, Martha Wayles, and Sally Hemings were half-sisters through John Wayles; the documented Jefferson controls descend through Thomas and Martha’s daughter Maria, who married her own first cousin, John Wayles Eppes; and the Carr and Jefferson lines intermarried as well. Deep shared ancestry among these families is the rule, not the exception, so real background relatedness is to be expected--relatedness that a single-relationship simulation would misattribute to the specific paternity claimed. Max Planck has now confirmed this concern on statistical grounds.

A question on the reference panel.

The scan for shared DNA draws allele frequencies from the 1000 Genomes GBR (British) panel, applied to admixed Hemings descendants. The authors report similar results with the CEU panel, but whether a European-only panel biases segment detection in admixed samples--and whether an African-ancestry or admixed panel would move the Eston-line result specifically--is a question worth a hard look.

A preliminary finding of my own, held tentatively.

In my own database, Madison Hemings descendants and the Carr nephews’ descendants share 10 to 20 IBD segments. If Thomas fathered Madison, essentially all of those shared segments should also appear in Thomas’s genome, since his only link to the Carrs runs through the parents he shares with the nephews’ mother, his sister. If a Carr nephew fathered Madison, only a minority--roughly a quarter--should appear in Thomas’s genome, the remainder coming from the Carrs’ non-Jefferson side. My data favor a Carr nephew over Thomas for Madison. I hold it tentatively, and there is a genuine tension in it--a Carr nephew sits two degrees further from Thomas, and truly long shared segments point to a closer relative, so if the study’s long Madison segments are real, a Carr father is hard to reconcile with them. But those long segments are exactly what the established method has not yet confirmed. Until we know which segments survive it, neither interpretation is settled; instead they are two sides of the same open question.

What would meet the standard.

Two things. First, release of the recovered genome and the underlying data, so the shared DNA calls can be reproduced with the field standard method rather than taken on the team’s own new pipeline. Second, the direct comparisons the standard has always required-Thomas’s DNA against Randolph’s for Eston, and against the Carr line for Madison. Until then, the fair reading is that their study raises the question well and recovers remarkable material, but has not yet answered it--and it should not be treated as settled while it remains an unreviewed preprint.