Methodology

General methodology

Archives, in person and online

The genealogical and historical research draws on primary records consulted directly in three countries. In Mauritius: the National Archives of Mauritius at Coromandel, the Civil Status Office, and the Mauritius National Library, for parish registers, manumission and notarial records, civil birth, marriage, and death registers, and newspapers and gazettes. In the United Kingdom: the National Archives at Kew, for Colonial Office correspondence, slave registration returns, and the compensation records created at abolition. In France: the Archives nationales d'outre-mer at Aix-en-Provence, for French-period notarial acts and administrative correspondence.

A single record is rarely treated as sufficient on its own. Pierre Jerome's 1812 manumission, for example, is corroborated by his own 1826 petition for his mother's freedom, Jean-Pierre Galichet's will, and later census and compensation records filed under entirely separate series. Wherever possible, the same event or relationship is checked against as many independent documents as survive before it is treated as established, and a fact-level correction on one page is checked against every other page where the same fact is restated, not just the page where it was first found.

Online research supplements the archive visits rather than replacing them: the UCL Legacies of British Slave-ownership database for compensation claims, cross-checked directly against the underlying Kew and Mauritius records, a process that has itself surfaced several confirmed errors in that database; online genealogy platforms, principally Ancestry, with additional leads from MyHeritage, for building and testing the family structure against the archival finds; and published secondary scholarship (see Sources & bibliography) for the historical and linguistic strands, where the aim is less original archival discovery than an accurate synthesis of existing research.

Confidence, stated plainly

Family history research constantly runs up against the limits of what can be known, especially for people who were enslaved, whose names were changed at will, whose families were separated, and whose lives were rarely thought worth recording in detail. Rather than smoothing over that uncertainty, every claim on this site is held to one of four levels: a measured fact, drawn directly from a document, test result, or record; a strong inference, well supported by the evidence but not itself directly documented; a hypothesis, plausible and clearly flagged as such; or family memory, oral history recorded faithfully, neither treated as literal fact nor dismissed. Where new evidence contradicts an earlier hypothesis, the hypothesis is revised rather than defended, which has happened more than once in the course of this research, including in the DNA methodology below.

The website itself

The site is part of the method, not just a way of presenting the results afterwards. Its four strands, genealogy, DNA, history, and anthropology, are cross-referenced throughout: a DNA segment against a documented migration wave, a surname against a slave-naming pattern, a Mauritian Creole word against its Borneo cognate. Building it that way, so that a single fact can be checked against everything else known about it, has surfaced errors that a linear document would likely have missed. The site began as a prototype built by my son Pascal; I have continued its development since.

DNA & genetic methodology

The full working, with every assumption stated and numbered, is kept as a standalone research note; what follows is the method in outline.

Starting from the measured DNA, not the family tree

There are two directions from which a family tree and a DNA test can be reconciled. The first, working bottom-up, sums each identified ancestor's expected contribution to predict a descendant's regional percentages, then compares that prediction to the measured test. Its weakness is that it depends entirely on having identified every contributing ancestor and knowing their composition: an unidentified ancestor contributes nothing to the prediction, so any shortfall is invisible, and there is no way to tell whether it reflects a missing ancestor or ordinary recombination noise.

This site uses the second approach instead, working top-down from the measured DNA, which is treated as ground truth, into the tree. An unidentified ancestor's contribution then becomes a calculated residual rather than an assumed input: the method never claims more knowledge than the data supports, and a gap in the family tree shows up as a quantified finding ("this much of a given region on one parent's side isn't explained by any identified ancestor") rather than as a silent gap in a prediction.

The four steps

First, one parent's contribution is established directly, here the mother's, from her own test result halved, and later confirmed independently by direct parental phasing. Second, the other parent's contribution follows by subtraction, the descendant's total measured DNA minus the known parent's half, a figure later measured directly as well, once phasing technology allowed it, and found to agree with the derived figure closely. Third, within each parent's known total, identified ancestors are matched to the regions their documented origin predicts. Fourth, whatever remains unallocated within a parental half is the measured signature of that half's unidentified ancestors, not a guess about who they might have been.

An ancestor's expected share

Every ancestor at a given generation contributes, on average, a fixed share of a descendant's autosomal DNA, halving at each generation back:

This is an expectation, not a guarantee. Because DNA recombines when it is passed down, the actual inherited fraction from any one ancestor scatters around the expected value, and the variance grows the further back the ancestor sits: at three generations back the real figure can range roughly 6–19% rather than sitting exactly on 12.5%, and by five or six generations back a given ancestor may contribute nothing detectable at all. Every single-ancestor figure on this site is therefore an indicative expectation, not a precise measurement, and is presented as such.

Four super-regions

For clarity, the many fine-grained regions a DNA test reports are aggregated into four super-regions matching the site's four narrative strands: the Indian subcontinent, Europe, Africa, and Austronesia (the Madagascar-to-Borneo signal). Most regional labels sort cleanly into one of the four. A few don't, and are handled explicitly rather than silently: Madagascar's population is itself a blend of Austronesian and Bantu ancestry, so a documented Malagasy ancestor's expected share is legitimately split across both the Africa and Austronesia columns; and at least one region reported by the testing companies sits in a genuine grey zone between Indian, East Asian, and Austronesian ancestry, and is assigned here to Austronesia on the strength of the family tree, which documents several Malagasy ancestors capable of producing that signal and no ancestor from mainland Southeast Asia. Where the DNA label alone can't settle a region's assignment, the family tree is used to break the tie, and that inference is flagged as such rather than presented as a direct reading of the test.

Assumptions, stated plainly

Beyond the ½ⁿ expectation and the super-region assignments above, two further principles govern how figures are attributed to specific ancestors. Where an ancestor's ethnic make-up is directly documented, their expected share is attributed to the corresponding super-region. Where it is not documented, it is left as part of the unallocated residual described in step four above, unless there is a specific, independent piece of supporting evidence, a surname's geographic distribution, a documented pattern elsewhere in the same family branch, or family memory, in which case it may be credited as a clearly flagged hypothesis, never presented as settled fact. Small percentages and trace regions carry real estimation noise and are not over-interpreted.

What makes the method defensible

The method starts from measured ground truth rather than assumption, and has been cross-validated two ways: the paternal half derived by subtraction and the paternal half later measured directly by phasing agree closely. It has also been replicated independently, a sibling's separate phasing reproduces the same maternal and paternal structure. Unknown ancestors are quantified residuals, not guesses, and every assumption that feeds into the figures is stated rather than left implicit. Most importantly, where the data has contradicted a working hypothesis, the hypothesis has been revised rather than defended. That discipline, not the striking conclusions it occasionally produces, is the basis for trusting the numbers.

The numerical results this method produces are set out on the DNA Analysis page; full citations for the tests themselves are on the Sources & bibliography page.