Method

A living person (80 or younger, with no death recorded) is named with their relationships and their town and country, and nothing more. No date, no record, no source, no note and no photograph; their page is not indexed, carries no structured data and is left out of the sitemap and the search index. The test is the estate's one shared rule: age by year (this year minus the birth year), a death or burial recorded means dead, and a citation from a public record source means the person is not treated as living. A person with no birth year cannot be shown to be 80 or younger; the archive's own statement that someone is living is accepted. Anyone who may be a child (under 18, or undated with a living parent) is not published at all: no page, and no name as anyone's relative.

Because this line reaches the present within three generations of Catherine Mary Sophia Booyzen, a living person is part of the descent: Cheryl Anne Lerena, David Defranceski's mother, and David himself, who is named on the home page with his own places by his own choice. Their siblings and other relatives are not named, and so have no page.

A removal request is honoured within days, without argument and without requiring a reason.

Unusually well. The Lerena archive began with almost nothing documented. This one begins with a MyHeritage tree of 15,643 people in which the Booyzen line is already carried back seven generations, and in which a large proportion of births, marriages and deaths already carry a FamilySearch citation to an original register.

That changes the job. The work here is not to find the family. It is to test what has been assembled — because a tree that good will be believed, and parts of it are wrong.

So this archive did three things the tree does not do:

• Listed every contradiction rather than choosing between them silently.
• Tested the family testimony against outside evidence — surname distribution, Cape Rebel district figures, the 1820 Settler record — instead of repeating it.
• Named what has not been searched, so the gaps are visible.

This is the thing this archive actually does, and until today it was not written down anywhere. It arrived by being wrong six times in a week.

An index that returns nothing is making one of two statements and it does not tell you which. Either the record does not exist, or your search could not have found it. Those are opposite conclusions and they look identical on the screen. So a negative is not publishable until something has been run through the same search that is known to be there — a control — and come back.

So the rule is: run a control, and publish it with the negative. When this archive reported that James Mountjoy is not among the discharge papers at Kew, it published alongside that ninety-five records for his regiment in the same series in a single page of results — because the number that makes the absence meaningful is not his, it is theirs.

The same search also turned up the reason the control mattered. The catalogue indexes the regiment as “21st Dragoons (Light)”, not as the newspaper wrote it and not as this archive had been writing it. One word order between a real record and a nil return.

A search that has not been controlled has not been done. It has only been attempted, and the difference between “searched and not found” and “could not search” is the most important distinction on this site.

Which is why What has been read marks a row void when the search itself was faulty, and blocked when it could not run — and why a build here now refuses if a blocked row does not say what would unblock it. A negative with no control behind it is an opinion with a citation.

This one was named on 21 September 2026 by the session keeping the D’Arcy archive, after the two of us had between us produced six of them in a day. It is the rule the controlled negative does not reach, because half of these never returned a nil at all. A test passed. A count came back. A confidence field said exact.

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.

The test that catches every one of them is the same, and it is one question: if the thing I am looking for were definitely there, could this instrument have told me? Ask a filter for something already seen with your own eyes on the page in front of you. Ask a comparison to fail on a difference you planted. Ask a generator to survive a rebuild, not a re-read.

None of them is a wrong answer. Each is a correct answer to a question nobody asked.

And one heuristic of this archive’s own did not survive the day. “Three identical failures is a fault; one is the weather” is sound for a machine under load and useless for a bug whose trigger is intermittent and whose mechanism is not. A build failure here was written off that way in the afternoon; it was a missing subshell that could have reported a killed build as build complete, and it had been there all along.

A failure that does not reproduce has not been explained. It has only been survived — and passing next time is evidence about the trigger and none at all about the mechanism.

One last thing, about why any of this got found. Eight of these faults were caught in a single afternoon, and not one was caught by an audit. They were caught because two archives kept describing their own work to each other in enough detail that the other could recognise the shape of a fault it had not gone looking for. Neither was checking the other. A wrong attribution, a comparison that could not fail, a trap that went silent, a stopper that stopped nothing — every one of them surfaced because somebody volunteered it first.

This archive already sets its negatives down at the same length as its finds, on the grounds that a family history publishing only its successes is not a history. The day added a second reason, and it is more selfish than the first: a fault written up as briefly as it is embarrassing is a fault nobody else can recognise in their own work. The length is what makes the shape visible from outside. That is a discipline rather than a courtesy.

And it is why the three tool faults needed somebody else at all. A claim about a family is a claim about the world, and the world can contradict it. A claim about an instrument is a claim about oneself — my filter filtered, my trap trapped, my stopper stopped — and nobody is a disinterested judge of their own competence. So the judgement never gets made, and the claim never reaches a boundary where anything would test it. Somebody describing their own instrument out loud is the cheapest boundary there is.

The second half of that sentence is the D’Arcy session’s, and it is the part that makes the rule usable: it says exactly what the re-run did prove, which is the thing a tired reader will otherwise take for a clean bill of health.

Six of these were found by measurement. The last two were found by checking an instrument against itself — asking the stale-index check about the file it actually reads, and asking the trap about the branch nobody had run. That is the half that generalises, because measurement only ever checks the thing you already doubted.

Which is why every gate added here is verified in both directions — silent on the real data, and made to fail on a fault planted on purpose. A check that has only ever passed has not been shown to be a check.

Nearly every genealogical search is a search for somebody you can already name. You type Hendrik Stefanus Booijzen, born about 1827, and the index tells you whether it holds him. That works, and it is how most of this archive was built. But it has a hard ceiling: it can only ever confirm what you already suspected. If the person you need was never written down by your family, you cannot type their name, and so you cannot find them.

The search forms will also take a relationship. Leave the forename blank, leave the district blank, and ask instead for any child whose mother's surname was Kolbe. The index then answers a question nobody in the family was in a position to ask.

Both results on this page's hardest day came out of that difference. Three person-searches for Elizabetha van der Merwe returned nothing, and the third of them — her son's own baptism — was the best test this archive could devise. One relationship-search on the Barrys returned two children on neither parent's death notice. The difference was not effort or luck. It was which question the form was asked.

The same logic runs underneath three other things this archive did. Reading a marriage register's Remarks column for the Master's estate number is a relationship search done on paper — it indexes from a living remarrying widower to a file about a dead first wife. Reading both parents' death notices rather than one gave a ninth child. Reading a whole surname out of a probate collection, instead of searching it for a person, gave twenty-nine.

The cost is honesty about noise. A relationship search returns the right family mixed with strangers, and it returns the same event indexed twice to two different villages. The two Barry names above are recorded as a lead and not as children, because an index line that fits a gap is not the same as a register that fills it.