The DNA

Three people in this family have taken the same test, and the results were updated to MyHeritage's v2.5 model this month. Set side by side they do something the registers cannot: they say what is in the body now, rather than what a clerk wrote down.

They also do something the registers never would — they disagree with each other, and with arithmetic. A child is exactly half of each parent. The percentages are not. That is not an error in the test and it is not a surprise in the family; it is what these models do, and this page is mostly about learning to read them honestly.

An ethnicity estimate is not a document. It is a statistical fit of one person's genome against reference panels of living people, expressed to one decimal place because the software can, not because the world is that precise. This archive grades every claim doc, inf or lore. These numbers get a fourth grade of their own — dna — and it ranks below doc, because a register entry names a person and an estimate names a probability.

Since 14 September 2026 there is a fourth reading on this page, and it is a different kind of thing. A Y-DNA haplogroup is not a statistical fit against reference panels — it is a mutation, present or absent, on the one chromosome that passes father to son intact. It is the only evidence in this archive that runs straight up the male line, and it is the only one that does not agree with the paper. It is at the Y chromosome below.

If these numbers behaved like inheritance, David's column would be the average of his parents'. Here is what happens when you actually take the midpoint.

The total drift is 42.4 percentage points. Every row moves. The largest movements are all in one direction: Balkan is 13 points higher than it should be, and the three British-Isles groups together are ten points lower — 23.3 expected, 13.4 measured.

Three rows land dead on the midpoint — Danish, South Italian and, to a tenth of a point, Breton. Two of Ivan's disappear entirely: his 4.0% Baltic and, from Cheryl, Sardinian and Swedish, all of which fall under the reporting floor once halved.

The reason is that these models do not blend, they assign. A stretch of chromosome is given to whichever reference population fits it best, and a person of mixed descent has long stretches that could plausibly go either way. Where the two sides of a family are far apart genetically — a Kvarner Croat and a woman of the Low Countries and the British Isles — the more sharply defined population tends to win the contested segments. Balkan is a tighter cluster than 'English', which is itself a blend of everything that ever landed there. So Balkan takes segments that a strict accounting would have split, and the British share shrinks.

Germanic goes the other way — 8.6% in David against a 2.6% midpoint, higher than either parent's contribution allows. That is the same effect running in reverse: Germanic sits geographically between his mother's Dutch and his father's East European, and it collects the segments that neither cluster claims outright.

Now the part that matters more than any percentage. Ivan's estimate contains no Italian at all. Not North Italian, not South Italian, not a fraction.

This archive's founding story is that the name came out of Carnia — the Val Degano in Friuli — and reached Istria by about 1679. If that is true, why is there no Italian in the father's DNA?

Because there was never going to be. Autosomal DNA halves every generation, and it halves fast:

The reporting floor on these estimates is around 1%. Read the table again with that in mind: everything older than Giovanni Battista, born 1787, is beneath the floor — and that is on the direct male line, which is the slowest-diluting path a surname takes.

So the Carnian origin is not contradicted by the DNA. It is simply out of range. An ancestor of 1679 contributes about one part in a thousand of a living descendant's genome, twenty times less than the test can see. The Omiš chart's fourteenth-century man contributes a number with four zeros after the decimal point.

This is the single most useful thing on this page. The paper reaches back to 1318. The DNA reaches back about six generations, to roughly 1790. They are not two views of the same distance. Where they overlap — Kvarner, generations 4 to 6 — they agree exactly. Beyond that overlap, the DNA has nothing to say, and an absence of Italian is not evidence against Carnia. It is evidence that the test works as designed.

The coloured continents are deep ancestry. The genetic groups listed beneath them are something else entirely — clusters of living testers who share recent ancestry in one place. They are a record of where a family was two or three generations ago, and they are blunt about it.

David carries four: South Africa #2, South Africa and Zimbabwe, South Africa #3, South Africa (Gauteng and Free State) — and England. His father carries none of them. Every one comes through his mother, whose own list reads South Africa, South Africa and Zimbabwe, and Argentina (Corrientes and Buenos Aires).

That is the maternal family's twentieth century written in genetic shorthand: the Low Countries and the British Isles by descent, southern Africa by residence — Gauteng and the Free State named outright — with a strand running to the Argentine littoral. The direct line's own migration is in the same list from the other side: David was born in Johannesburg, and his father came to it from Senj.

These groups do not date anything and they do not name anybody. What they do is tell you where to write next — and on the maternal side this archive has written almost nothing.

One of those groups now has a person behind it. Cheryl's father — Nuno Fernando, called Larry — was born in South Africa, and his father came from Argentina while his mother came from England. That is a documentary explanation for two things the estimate had only guessed at: the Argentina (Corrientes and Buenos Aires) cluster, and a share of the English. A genetic group is a question; this is the first time a family document has answered one.

Ivan's estimate carries 10.3% Greek and Albanian, a tenth of him, and it survives into David at 2.5%.

In the eastern Adriatic this is not exotic. It is partly the deep Balkan substrate that every population on that coast shares, and partly something more specific: for four centuries the Venetian Stato da Màr moved Greek and Albanian people up and down this sea — the Arbanasi settled outside Zadar in the 1720s, Greek communities held churches in Venice, Trieste and the Dalmatian ports, and Venetian galleys carried crews from every shore they touched.

This archive already has a family out of Crete at Seget. That is a different family and no connection is claimed — but it is worth saying that a Greek-and-Albanian tenth in a Kvarner man is exactly what four hundred years of Venetian sea traffic would leave behind, and the registers on At sea are full of the ships that did it.

The same test publishes how often the haplogroup turns up among its own testers, by the country each tester names as the origin of their direct paternal line. This is the whole of the top of that table, and then the two rows that matter to this family.

Italy is on the list. Five men out of six thousand, but they are there, and that matters more than the headline does: this lineage is rare on the Adriatic, not impossible on it. Croatia is not on the list at all — which means no Croatian tester in the panel has reported it, not that no Croatian carries it.

The archive states the rarity and refuses the leap. Rare is a probability. Impossible would be a claim, and nothing here supports one.

This is not a Y-DNA test. MyHeritage reads the haplogroup off the same autosomal microarray chip that produced the ethnicity percentages. Those chips carry a limited and uneven scatter of Y markers. That is enough to place a man on a major branch of the tree; it is not enough to place him on a twig, and it is not the same instrument as a dedicated Y test.

Two kits agreeing is not two tests agreeing. David and his father return the same call, and if both kits ran on the same chip a systematic error would produce exactly that agreement. (This objection was written in the morning. A second company answered it the same day — see the second opinion below.)

Nothing here identifies a generation. A haplogroup is a fact about a lineage, not an event with a date. This page does not claim a break in the male line at any point, does not name one, and has no basis to. What it reports is a marker and an inconsistency with expectation — and the honest position is that the inconsistency has more than one explanation, including a rare-but-real European L-M27 of the kind those five Italians carry.

The part that would say when is behind a paywall. The Time Tree — the branch structure and the dates on it — is locked, and unlocks only with a Big Y-700 from FamilyTreeDNA.

The test matches the lineage against published ancient genomes, and for L-M27 it returns two — both from the same place, and it is a place worth naming.

Areni 44 (4330–4056 BCE) and Areni 46 (4230–4000 BCE), both male, both from Areni, Armenia, both catalogued as Copper Age Caucasian. Areni 44 carries Y-DNA L-M27 and mtDNA H2a1; Areni 46 carries L-FTA37789 and K1a8*. They are samples I1634 and I1632, published in Lazaridis et al. 2016 and carried forward into the Allen Ancient Genome Diversity Project and Akbari et al. 2026.

Areni-1 is the cave that produced the oldest known leather shoe and the oldest known winemaking installation in the world. The shared paternal ancestor is placed at about 6500 BCE.

That is eight and a half thousand years ago and it says nothing whatever about this family. It is deep prehistory, and it belongs on this page as what it is — the furthest back a single unbroken thread of this line can currently be followed, and a genuinely published pair of skeletons at the end of it.

The same feature offers a notable connection: Sayyid Ajall Shams al-Din Omar al-Bukhari (1211–1279), the first provincial governor of Yunnan, whose descendants include the Ming admiral Zheng He, and whose male-line descendants were shown to carry L1a-M76 in a 2013 study arguing a southern Persian origin for the family.

The test rates the connection rare — 1 in 782. The shared paternal ancestor is placed at 13,000 BCE.

Thirteen thousand BCE is seven thousand years before the Areni men and about twelve thousand years before anybody wrote anything down. This archive already made this argument once, at length, about the ancient-genome scores further up this page, and the conclusion is the same one: a shared ancestor in the Palaeolithic is a fact about the species, not about a household. The feature itself calls these «fun facts», which is the most honest label on the whole screen.

It is recorded here so that nobody has to go looking for it twice.

«David, your paternal haplogroup is L-M27.»

23andMe publishes the whole path it walked: A → A1b1-L419 → BT-M42 → CT-M168 → CF-P143 → F-M89 → GHIJK-M3658 → HIJK-M578 → IJK-L15 → K-M9 → LT-P326 → L-M20 → L-M22 → L-M2481 → L-M27.

It also publishes a change log, and one entry in it matters here: on 22 April 2024 the company «corrected variant metadata errors and excluded variants that were performing poorly on our v5 chip», specifically so that «some customers received more accurate paternal haplogroup assignments». The obvious failure mode — a bad Y marker on a microarray — is the one they went in and cleaned up.

One caution survives, and it is a real one. If David's MyHeritage kit is an upload of this same 23andMe file, then the two readings are one reading counted twice. This archive does not know which, and is asking. What is certain either way is that his father's kit is a separate sample of a separate man, and it returns L-M27 as well.

Haplogroup H. David's mother's mother's mother's line, back to a woman who lived about 18,000 years ago, by way of L → L3 → N → R → H.

H is the commonest maternal lineage in Europe — roughly four Europeans in ten carry it. It says the maternal line is European and it says almost nothing else, and this page would rather record that plainly than dress it up.

An AutoCluster is not another estimate. MyHeritage takes the matches in a band of shared DNA, tests every one of them against every other, and puts the people who all match each other into a square. A square is usually one couple's descendants, seen from the outside: you do not know who the couple were, but you know the people in the square share them.

Two reports were run on 15 September 2026 — David's kit and his mother's. His returned 73 matches in nine clusters, from 40.0 to 130.2 centimorgans. Hers returned 103 matches in fourteen clusters, from 50.6 to 332.8. And the first thing to say about them is the thing the registers have never been able to say: three of David's nine clusters are Croatian, and not one of his mother's fourteen is.

The three Croatian clusters hold twenty people and 928 centimorgans between them. The largest is twelve people sharing 560 cM. Their surnames belong to the Kvarner and to Lika — the coast and the hinterland behind Senj — and three of the twelve carry the Americanised spellings that mark the emigration to the United States, which is the same road this archive has been reading in the Croatia filed abroad collections. The two smaller ones are four people each, and their surnames are Istrian and Kvarner again.

Where the cross-check is strong, and where it is not. The two reports do not start at the same place: his mother's stops at 50.6 cM and his goes down to 40.0, so most of the twenty Croatian matches are below the floor of the report they are being checked against. Only two of the twenty are above it. Both are absent from hers. That is consistent, and on its own it would not be enough.

What carries the weight is the paper. All four of David's paternal great-grandparent lines are Croatian — Defranceschi at Ledenice and Crikvenica, Kovačina at Crikvenica, Blažević at Rijeka and Senj, Papić at Senj — and there is no Croatian ancestry on his mother's side at all. A Croatian cluster in his kit has one place it can come from.

Cluster 9 is the honest failure. Three people, Afrikaans surnames, nobody in his mother's report — but all three sit below her floor, so her report could not have held them whether they belong to her side or not. It is unplaced, and it is listed as unplaced rather than counted on either side.

What would settle it in an afternoon: run the same report on his father's kit. Ivan tested too, and his kit is still on the account. An AutoCluster of his matches would put the Croatian clusters on the paternal side by arithmetic instead of by inference, and it would name the groups this page can only count.

Each match in a cluster may have posted a family tree, and most of these have. Reading the trees behind the three Croatian clusters does something the cluster numbers cannot: it says where.

The cluster of twelve is Senj and the Vinodol. The ancestral surnames standing behind its members, in their own posted pedigrees, are Pavelić, Vukelić, Biondić, Lopac, Hečimović, Prpić, Krmpotić, Butorac, Balen, Rukavina, Pastuović, Šojat and Blažević. That is not a scatter of Croatian names. It is one population: the Bunjevci of Senj, Krivi Put, Krmpote and the Vinodol, the villages on the slope behind the town. Krmpotić is named for Krmpote, which is the next valley to Ledenice — where Anton Defranceschi was born in 1890. Biondić is a Senj name outright.

The cluster of four beside it is Lika, the plateau behind that slope: Račić, Orešković, Matasić, Sekulić, Rajković — and Blažević again.

The third is the islands — Babin, Hromin, Škopić, Vlakić, Jadrijev — carrying the by-name formations (Ivanov Babin, Hromin Škopić) that belong to Krk and the northern Adriatic islands.

Blažević is the name to stop on. David's paternal grandmother was Hedviga Blažević, born at Senj in 1926; her mother was Milka Lucia Papić, born at Senj in 1886. Blažević stands in the trees of two separate Croatian clusters. The registers put this family at Senj and in the Vinodol; the DNA puts it in the middle of the population that lives there. Those are two different kinds of evidence arriving at the same village.

What this is not. A surname in a match's tree is not a proven common ancestor. Nobody here has been joined to anybody by a record, and none of these clusters has been traced to a couple. What is shown is that the pool of names behind David's Croatian matches is the pool of names in the parishes this archive has been reading — which is the correct amount of claim to make from it, and it is graded dna, below doc, for that reason.

Of the four AutoCluster reports read on 16 September — David's, his mother's, his father's and a third on the mother's side — exactly one match in any of them carries this surname, and it is in his father's.

It is Gary DeFranceschi of Brooklyn, and he is not a stranger: he is already in David's family tree. What was a Smart Match is now a blood tie, because his kit matches Ivan's.

The paper and the DNA arrive at the same man. Read both male lines upward and they meet exactly:

David → Ivan (1951, Senj) → Ivan (1925, Crikvenica) → Anton (1890, Ledenice) → Josephus (1863, Gračišće) → Antonius (1825, Gračišće)

Gary → Joseph Jordan (1929, Clifton) → Giuseppe (1892, Gračišće) → Antonius (1856, Gračišće) → Antonius (1825, Gračišće)

Josephus of 1863 and Antonius of 1856 are brothers. Their father is Antonius Defranceschi, born 1825 at Gračišće, whose own father is Giovanni Battista of Gologorica, born 1787, of house 5 — which is the spine of this archive, steps two and three.

So Ivan and Gary are third cousins, and the arithmetic of that is the last proof. Third cousins share about seventy centimorgans, and a son inherits roughly half of what his father has. Ivan's report has a floor of 40 cM and Gary is in it. David's has the same floor and Gary is not — which is exactly where half of seventy falls. The match is in the father's report and absent from the son's for the reason the pedigree predicts.

One branch went from Gračišće to Brooklyn and one went to Ledenice and then to Johannesburg, and a hundred and thirty years later the DNA still knows.

Ivan's report existed all along. Eight AutoCluster reports came off this account on 15 September 2026 and only five were ever opened; his was one of the three that sat unread. It was found on 16 September and read the same hour.

94 matches in thirteen clusters, and the first of them holds fifty people. Every one of the twelve people in David's Croatian cluster is inside it, and so is one of the four from his next Croatian cluster. Seen from the father, those are not three clusters. They are one.

And now every cluster David has is decided by arithmetic rather than by surname.

Clusters 2, 5 and 6 — the three Croatian ones — have ten, three and two of their members in his father's report and not one in his mother's. Clusters 1, 3, 4, 7 and 8 are the other way round: members in hers, none in his. There is no cluster with members in both.

This morning [the surname pool](#autoclusters) said the three were Croatian and [a third kit](#autoclusters) said a maternal relative was absent from them. Both were consistent and neither was proof. This is proof. The clusters carrying Pavelić, Vukelić, Biondić, Hečimović, Krmpotić and Blažević descend from David's father, and therefore from Senj and the Vinodol, because that is where his father's family is from.

Cluster 9 is still unplaced and is still called that. Its three people are in neither parent's report, and all three sit below both floors, so neither report could have held them.

A fifth report came off the account on 15 September and nobody had opened it: a third kit, belonging to a relative on David's mother's side. Read on 16 September it does for her fourteen clusters what Ivan's kit would do for his three.

It shares 31 matches with David's mother and only 6 with David. That ratio is the whole argument: a child inherits about half of what a parent has, so a relative who touches the mother's list five times as often as the son's is on her side and not his.

And it splits her fourteen clusters nine to five. Nine of them contain somebody the third kit also matches; five contain nobody at all. 81 of her 103 matches are in a cluster the third kit shares, 22 are not. The name on that kit is her own mother's maiden name under its other spelling, which is what the nine clusters are: her mother's family. The five it never touches are her father's.

The by-product is the useful part for this archive. Against David's own nine clusters the third kit touches 1, 4 and 8 — and not 2, 5 or 6. Those three are [the Croatian ones](#autoclusters). A third kit, tested independently, puts a maternal relative in three of David's clusters and in none of the Croatian three. That is no longer an inference from a surname; it is arithmetic on a third person's DNA.

Ten of the twelve people in the Croatian cluster of twelve have posted a family tree, and on 16 September all ten were opened. The question was whether they name the couple they descend from.

They do not, and the reason is honest rather than disappointing. Most of these trees are shallow — three of them show three, seven and twelve people. Only two are deep, at 3,597 people and 1,521. The two deep ones do not touch each other on a single named person.

What they share is a pool of surnames, and it is a tight one.

Vukelić stands in three independent trees — which is the strongest signal any name in this cluster gives. Prpić stands in two. Behind them: Blažević (Antonia, 1852–1930, whose grandfather was Anton «Kljapo» Blažević, 1824–1897), Pavelić, Krmpotić, Tomljanović, Biondić, Lopac, Katalinić, Rončević, Šojat, Butković, Šolić, Delale.

Every one of those is a Senj-hinterland name — the Bunjevci of Krivi Put, Krmpote, Sveti Juraj and the Vinodol, the villages on the slope behind the town Hedviga Blažević was born in.

So the next name to chase is Vukelić, not Blažević. Blažević is the one this family carries and therefore the one the eye goes to; Vukelić is the one that recurs across the most trees that have nothing else in common. No living match's name is published here, and none will be.

The same test offers a list of notable individuals whose ancient DNA has been sequenced, ranked by a Genetic Fit score. The scale runs the wrong way round: lower is closer. Under 1.0 is «very close», 1.0–3.0 «close», 3.0–5.0 «moderate», above 5.0 «distant».

The temptation is obvious. King Béla III of Hungary, Croatia and Dalmatia sits at the top of David's list at 2.196, the best score anyone in this family has, and it is very easy to read that as a line of descent from a medieval king.

It is not one, and the three lists prove it themselves.

If the score were meaningless, the two parents' lists would be interchangeable. They are not, and the differences fall exactly where descent says they should.

Philip Calvert, an English colonial governor of Maryland, appears on David's list and his mother's — and not on his father's. She is the parent with 13.7% English, 14.7% Irish and 18.2% Scottish and Welsh. Jean-Paul Marat appears on hers alone; she is the parent with 5.1% French.

Running the other way, the two Rurikid princes — Gleb Svyatoslavich and Izyaslav Ingvarevich, of the house that ruled Kyiv and Novgorod — appear on David's list and his father's, and not on his mother's. He is the parent with 20.8% East European and 4.0% Baltic. King Ladislaus I appears on his father's alone.

So the method is doing something real at the level of regions. It is only when it is read as a family tree that it collapses.

Read the fourteen names again and notice what is not there. There is no Italian on any of the three lists. No Croatian. No Venetian, no Istrian, no Friulian, nobody from Carnia or Vodnjan or Senj.

The list is Hungarian and Transylvanian nobility, princes of the Rus', a Saxon architect, a Scanian bishop, a German composer, a Bronze Age battlefield in Mecklenburg, an English colonist and a French revolutionary.

That is not a statement about this family. It is a statement about which graves have been opened and paid for. Ancient-DNA projects have concentrated on Hungarian royal tombs, Rurikid burials, and a handful of famous individuals; the medieval dead of the eastern Adriatic have largely not been sequenced. A list of notable individuals is a map of published research, and this family's own coast is a blank on it.

It is the same rule as the rest of this page, in a new dress: an absence in the data is a fact about the data.

One thing is worth keeping, stated for what it is — a resonance, not a relationship.

The name at the top of David's list, Béla III, was king of Hungary, Croatia and Dalmatia. Ladislaus I, on his father's list, is the king who took Croatia in 1091 and founded the bishopric of Zagreb. And John Corvinus, on all three lists, was Ban of Croatia and Dalmatia at the end of the fifteenth century.

Three of the fourteen held the crown or the office that governed the shore this family lived on. Senj, where Ivan Defranceski ended up, was a royal free city of that kingdom; the Vinodol coast below it was held from the same crown.

That is not descent. It is something quieter and, for an archive that keeps a sovereignty chart, more interesting: the population of the Carpathian basin and the Croatian littoral has been one interbreeding world for a thousand years, and a genetic profile from it will land near the people who ruled it — because they came out of it too.