Method
Every claim on this site carries its confidence. A thing that is documented says so; a thing that is family tradition says that too, and is kept anyway, because pretending otherwise is how family myths become family history.
The five words
All seven of these family archives grade a claim with the same five words, so a level means the same thing wherever you meet it. The word is always printed next to the colour, because a reader should never have to know the palette to know the confidence.
- Documented
- A record was read, and it says this. The source is named.
- Probable
- The evidence points one way and nothing found so far points the other. Not proven.
- Inferred
- Reasoned from what is known — a gap of the right size, a name in the right place — rather than witnessed in a record.
- Family
- Told, remembered, or written on a family chart. Kept, because discarding it would lose the lead; labelled, because it is not evidence.
- Disputed
- Two sources disagree and neither has been ruled out. Both are shown.
A great deal of this archive's medieval evidence is not the document. It is a calendar: a summary of a deed, made by an archivist who had the parchment in front of them, and printed in a catalogue. Nearly everything on the name in its own century is of that kind.
It sits between documented and inferred and this archive grades it documented, for one reason: a calendar is a reading by somebody competent, working from the original, whose job was to describe it accurately — which is a different thing from a conclusion. What it is not is the deed. A calendar can mis-transcribe a name, compress a clause, or date a document wrongly: Berkeley Castle dates BCM/A/1/24/173 to the fifteenth century and the printed translation of the same transaction dates it 1329, two hundred years apart. Where a page rests on a calendar it says so, and where a calendar and a printed text disagree, both are shown.
Some pages carry this instead. It states the thing a reader is likely to assume — the connection that would be tidy, the identification that almost works — and says plainly that this archive does not assert it. An archive is judged as much by what it refuses to say as by what it says.
How a person here gets graded
The five words above are not applied by hand, person by person. Every one of the 473 people in this archive is graded from provenance.json, which records what has actually been read:
- Documented — at least one record names the relationship: a Queensland registration, a baptism, a passenger record, a sworn deposition. 33 people.
- Disputed — they hang off one of the two grafts. 71 reach the archive only through the Hornby claim, a single unevidenced link; 27 only through the Keele link, which was looked for in the right registers and found to be wrong. 98 people.
- Family — on the family tree, with nothing found for them either way. 342 people. This is most of the archive, and saying so is the point.
Which means the proportion is visible rather than implied: seven per cent of the people here are proved by a record. The rest is a tree that has not been tested yet, and a hundred people hanging off two links that have.
The five labels
| documented | A named source with a reference, and where possible the scan itself. |
| inferred | A reasoned conclusion drawn from documented facts, with the reasoning written out in full so that it can be overturned by anyone who finds it wanting. |
| superseded | Asserted in the family record, and now displaced by a document that says otherwise. |
| disputed | Asserted in the family record but unsupported, or contradicted, by what can be seen. |
| family lore | Told, remembered, not corroborated. Kept because it is precious; labelled because pretending otherwise is how family myths become family history. |
Living people
Living people are omitted from this build entirely — not hidden, not gated behind a login, not present in the output at all.
The rule is applied once, in code, at the boundary between the research data and the published site — so a person cannot be living on one page and dead on another.
Judging people only by their own dates does not work on a tree this size, because thousands of them have no dates at all: medieval nobility, unnamed spouses, half-remembered great-aunts. Treating all of those as living is safe but useless — it would withhold the fourteenth century. So instead the build computes, for every person, an upper bound on their birth year: the latest year they could possibly have been born, given their own recorded events, their children's births, their marriages and their parents' bounds. Somebody is published only when even that latest possible birth is more than a hundred years ago.
Every step of that reasoning is a genuine bound rather than a guess — a parent is at least fourteen years older than their child and at most sixty; a person existed at their own recorded events. A recorded birth is never overridden by inference, because an inference that made a living person look older would publish them. Where nothing whatever is known, the person is presumed living.
On this build the rule judges 2,034 of the 15,643 people in the tree to be living, and removes 4 from the ancestor set. Where a living person sits between two published generations, the line is shown as continuing but the person is not named — and a living spouse, parent or child is not named on anybody else's page either.
A removal request is honoured within days, without argument and without requiring a reason.
And the rule is checked, not just stated
A rule applied in code is only as good as the proof that it held. Every build now runs tools/check_living.py, which reads the built HTML — the only place a leak can actually be seen — and refuses to ship if any of the 1,768 name phrases belonging to a living person appears on a page. It fails the build rather than warning, and it fails the build if it cannot read the tree at all, because a guard that cannot check has not passed.
It does not work from a list of living people, and that is deliberate. A file naming two thousand living relatives, committed to a public repository, would be exactly the disclosure this rule exists to prevent. The names are read at check time from the GEDCOM, which is never committed. What is committed is the declaration: the policy, and the exceptions.
Where the data comes from, and what that means
The spine of this archive is a MyHeritage family tree of 15,643 people, exported to GEDCOM and parsed here. That is a starting point and not an authority. An online tree records what somebody believed; this archive's job is to say which of those beliefs survive contact with a document.
So the tree is treated as a witness rather than a source. Where it carries a citation, the citation is shown on the person's page and marked as attached in the tree rather than checked here. Where this archive has gone to the original, the page says so.
How to read a pedigree chart on this site
Every person page opens with one, and the border down the left edge of each card says how that relationship is known — nothing else. It is the only thing the border encodes, because conflating "this is my ancestor" with "this is proved" is how a pedigree starts lying.
| The border | Means | The card also says |
|---|---|---|
| Solid | the link is written down in a record this archive has opened and read | which record — a registration number, a file reference, a deposition |
| Dotted | the link comes from the family tree and has not been tested | tree |
| Dashed, name struck through | the link comes from the tree and sits above one of the two joins this archive has tested and rejected | which graft — Hornby or Keele |
A correct answer to a question nobody asked
This archive's oldest rule is that a null must carry a control — a test that could have failed and did not. It is written entirely about zeroes, and on 21 September 2026 it became clear that is only half the problem. Three faults were found here in one day and none of them returned a zero.
- A sweep of a catalogue returned a hundred records and printed the true count of a hundred and twenty-six beside them. The number was right and it was the number of records fetched, not the number that existed.
- A spelling table returned 371. That was the count of index entries, and the page called them baptisms; one row in six is a repeat.
- A map said a pin was exact. The field recorded that the lookup had succeeded, not that it had answered the question, so Berkeley spent weeks on the geocode for the word England with six people on it.
An instrument can be working perfectly and be pointed at something else. What it reports is then true, and about the wrong thing, and it arrives wearing exactly the face of an answer. A nil at least announces itself as bad news worth checking. A confident yes does not, which is why each of the three above survived a commit message, a published page and, in one case, a gate.
So the rule is put a second way, and this one can be run rather than admired: if the thing I am looking for were definitely there, could this instrument have told me? Ask a filter for something already visible on the page in front of you. Ask a comparison to fail on a difference you planted. Ask a generator to survive a rebuild rather than a re-read. The operational form is the Booyzen archive's, arrived at from three faults of its own the same afternoon — a comparison that iterated dictionary keys and could not fail, a place filter returning nought for a place printed in its own results, and a confidence field reporting on the wrong thing.
Three of this archive's faults were found that way in a single afternoon, and all three were in its own tools — a stopper that selected nothing, an exit trap that reported a clean pass over a build that had been killed, and a check asked about a file it does not read. That they were all tools is not chance. A claim about a family arrives from outside and is graded on the way in; the world can contradict it. A claim about an instrument is a claim about oneself — my filter filtered, my trap trapped — written by the one person who needs it to be true, and it crosses no boundary where anything would check it. The boundary on 21 September was a neighbouring archive describing its own work in enough detail that the shape was recognisable from here. That only happens if both sides write up what they got wrong at the same length as what they found.
Corrections
A correction is the most valuable thing anyone can send this archive. If something here is wrong — a date, a place, a join, a person who should not be published — say so, and it will be changed and the change recorded.