What was taken back
Every claim this archive has published and then withdrawn, in one place, with the cause named. Nothing here has been quietly edited away: each of these sentences was on the site, and some of them were on it for weeks.
Why this page exists
A family history that only shows its conclusions is asking to be believed on trust. This one has been wrong nineteen times, and the corrections are more useful than most of the findings, because they are the only part of the site that says anything about how the work goes wrong.
The right-hand column is the point. Sort by it and a pattern appears that no single correction would have shown.
A document read past what it says
A document read wrongly, or a conclusion drawn past what the document says.
The archive's own working
A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
A part searched, reported as the whole
A search run over part of the ground and written up as though it covered all of it.
A conclusion that does not follow
A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
The tree repeated before it was tested
A claim taken from the family tree and repeated before it was tested.
A fact cited back to itself
A fact derived from a reference, written into the record, and then cited back as evidence.
A list mistaken for a population
A negative built from a list mistaken for a population — the archive searched what it happened to have and reported the answer as though it had searched everything.
The list
was ““25 records and genuinely filtered — every one has a Kolbe parent””
now FIFTEEN OF THE TWENTY-FIVE. The other ten are KOLOBE, KOLKOBET, COLBERT and CULBERT — Sesotho and other surnames reached by FamilySearch's fuzzy name matching, which is the index working as designed. THE FILTER IS REAL AND THE NAME IS FUZZY, and those are different claims. q.fatherSurname does restrict to records having a parent of that surname; it does not restrict to the spelling you typed. HOW IT HAPPENED, WHICH IS THE PART WORTH KEEPING: the result rows were scraped with a /kolbe/i test that returned fifteen, and the page's own total of twenty-five was then reported as though all twenty-five had been looked at. Counting the hits and quoting the corpus are two different measurements and only one of them was made. Caught within the hour by going back to read all twenty-five rather than the matching ones — which is also what turned up an eleventh child for the Commandant.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Three pages carried the John Barry conflation as unresolved for three weeks after it was resolved”
now THE WORK WAS DONE ON 14 SEPTEMBER AND ONLY THE PERSON'S OWN ENTRY WAS TOLD. The Yorkshire John Barry was separated from the Cape one and followed past the date the Cape one died — the 1851 census has him a SHIP OWNER at Ruswarp, Whitby, the only man of that age and name in the county, with Whitby deaths registered in 1858 and 1860; and the Bilsdale burial he was supposed to have is absent from 5,934 indexed Bilsdale burials. Work list row 83 was closed the same week. MEANWHILE THREE PLACES WENT ON SAYING THE OPPOSITE. /tree/ carried a panel ending “until they are separated nothing above it can be trusted”; /disputed/ graded the dispute FATAL with the instruction “Separate them”, on a page that has a Resolved status and uses it elsewhere; and the ancestor chart labelled him “John Barry, 1796–1853”, which is the conflated pair — 1796 is the Whitby man's birth year and the Cape man has none. THE TEXT LIVED IN THREE PLACES AND THE TRUTH IN A FOURTH: a hardcoded panel in tree.astro, a wall node in pedigree.json, a row in disputed.astro, and the finding itself in people.js. Correcting one corrects nothing. All four now agree, and the person's birth field says plainly that 1796 belonged to the other man. THIS ONE REACHED READERS, unlike the latent contradiction corrected earlier today: /tree/ and /disputed/ are both linked from the main menu.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “MHG 1095/48 reported as still needing to be ordered from Pretoria”
now IT WAS READ A MONTH AGO AND ITS IMAGES ARE ON THIS SITE. The death notice of George Downing Mountjoy Booyzen is transcribed at /documents/ with three images, and it is the document that corrected his death date to 11 December 1947 — 21 February 1948 being the day the notice was signed. THE MISTAKE WAS METHOD, NOT READING. Today's sweep tested every item on the held Pretoria order against FamilySearch, found this man indexed only as the Spouse on his wife's 1944 file, and concluded from that alone that his own estate “stays on the order”. Searching one outside index is not the same question as asking what this archive already holds, and only the second question was skipped. THIS IS THE THIRD INSTANCE TODAY AND THE SECOND AFTER THE RULE WAS WRITTEN. A correction filed this morning named the same habit; row 212 was then given a rule saying each item is to be checked against this archive BEFORE it is opened; the rule was applied to the six Mountjoy estates and not to the Pretoria order an hour later. THE SOURCE REGISTER WAS ALSO CARRYING IT BOTH WAYS and would have caught me: the estates table chipped the file Blocked while the read table vouched for the same document with its image published. Both corrected. A gate was considered and refused — only two references in the whole register are vouched by the read section, so a check would be measuring a population of two.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was ““Not one of the ten is joined to this house yet”, written of the Mountjoy estates on 27 September”
now TWO OF THE TEN WERE ALREADY DRAWN AS CHILDREN OF THAT HOUSE, AND HAD BEEN FOR SEVENTEEN DAYS. Estate KAB MOOC 6/9/5047: 58379 was read and published on 10 SEPTEMBER, and its distribution account names the heirs of James Jonathan Mountjoy of Barkly East. ANNIE LOUISA NEPGEN, BORN MOUNTJOY and LOUISA ELIZABETH DIESEL, BORN MOUNTJOY are two of them, and both appear on this site’s own /households/ page as his daughters. The NAAIRS sweep of 27 September then listed the same two women among ten estates it called unattached. THE SENTENCE WAS NOT STALE, IT WAS WRONG WHEN IT WAS WRITTEN — the search result was written up without being checked against this archive. THAT IS THE THIRD INSTANCE OF ONE HABIT. Work list row 212 did it again on 8 October, listing an estate image to open that question 12 already records as READ; and a letter of 20 September asked a paid researcher for a death notice “for the parents” of a woman whose parents were already held here. The failure is never the searching. It is writing down what a result list says without asking what this archive already holds.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “James Montjoy's death notice cited as image 338 of film 007844936”
now IT IS AT IMAGE 339, AND 338 IS ANOTHER FAMILY'S DEATH NOTICE. Both frames were opened on 9 October: image 338 is the death notice of CHRISTOFFEL HENDRIK OLIVIER, trader, aged 56 years 4 months and 10 days, who died at his own house in the town of Graaff-Reinet on 15 June 1840 — his parents, not he, are the Cornelis Hermanus Olivier and Agnieta Elizabeth van Wyk on line 3, and this draft said otherwise for ten minutes before line 1 was read. Image 340 is WILLEM JACOBUS VAN DER MERWE of Bouwbosch Fontein. NO CONNECTION TO THIS ARCHIVE'S OLIVIERS IS CLAIMED OR LOOKED FOR; the name is here only to prove the frame is not Montjoy's. The James Montjoy notice, stamped 4393, is at 339, which is where the FamilySearch record ark 1:1:QP4Z-WK68 lands. THE READING OF THE DOCUMENT WAS NEVER IN DOUBT — every fact this archive published off it is confirmed: saddler, born in “Europe”, parents NOT KNOWN, a house and erf at Cradock worth £90, died in the village of Cradock in September 1827, and four children named James, George, Johanna and Jacoba, all minors. WHAT WAS WRONG WAS THE WAY BACK TO THE PAGE. A citation off by one sends the next reader to a stranger's family, and this one was published in THREE places — /sources/, question 12, and /documents/, which also sets out the arithmetic that produced it. THAT ARITHMETIC IS THE LESSON AND IT IS BEING KEPT: the reference numbers do not ascend by exactly one per image, because an estate running to two or three frames breaks the step, so interpolating from image 330 at reference 4386 predicts 4393 at image 337 and the narrowing stopped one frame short. An interpolated citation is a PREDICTION and has to be re-read off the frame it landed on.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Three Booyzens listed together as deaths in concentration camps”
now ONLY TWO OF THE THREE DIED IN A CAMP. Read on the University of Cape Town database's own pages rather than through two intermediaries: Gert Jacobus (Johannesburg RC, aged 2, farm Rietvallei, district Marico) and Gertruida Magdalena (Vereeniging RC, aged 19, farm Rietvlei, district Heidelberg) are each recorded “died in camp: YES”. JAN GERHARDUS LODEWICUS, AGED 2, IS RECORDED “DIED IN CAMP: NO”, with no camp and no farm attached at all. Listing the three together was this archive's compression, not the database's — and it is the same failure as reading an index line as though it were the record. Gertruida's entry also carries the note “or Irene RC”, so even her camp is uncertain.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The National Archives' catalogue given as “national.archives.gov.za/naairs.htm””
now THE LIVE DOMAIN HAS NO DOT IN IT — nationalarchives.gov.za. The host as published has no address record and no mail record at all. It appeared in the note of 13 September that ends “written down because the wrong address cost this archive a day”, which is the row that exists to stop exactly this. Found by a tool written to check whether the sources this site cites are still there; it looks for hosts named in PROSE as well as in links, because this archive writes its sources in sentences.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was ““WILLEM HERMANUS PIETERZEN — Anna Catharina Slier's FIRST husband””
now HE IS THE SON. De Villiers's volume 3 gives “HERMANUS PIETERSEN, namen van ouders onbekend, gehuwd 5 April 1807 met ANNA CATHARINA SLIER”, and their children as “1 Willem Hermanus, gedoopt 12 Maart 1809; 2 Johan Christiaan, gedoopt 28 April 1810”. THIS ARCHIVE BUILT ONE MAN OUT OF TWO DOCUMENTS THAT SPELL HIM DIFFERENTLY — the 1817 marriage register's HARMANUS PIETERZE and the 1840 death notice's WILLIAM PIETERSEN — and a long name looked like a way to reconcile them. The husband is Hermanus; Willem Hermanus is the boy baptised in 1809. This archive half knew, because its own note already called the 1851 godfather “his son”. THE NAMING CLAIM SURVIVES AND IMPROVES: Willem Hermanus is Willem for his mother's father, Willem Slier of Hamburg, and Hermanus for his own father. A SECOND HALF-BROTHER, JOHAN CHRISTIAAN, WAS INVISIBLE FOR AS LONG AS THE TWO MEN WERE ONE.
how A document read wrongly, or a conclusion drawn past what the document says.
was ““LUUS with BOOYZEN returns NOTHING” — published against the five unplaced Silverton children”
now IT RETURNS A MARRIAGE. The Marriage Transcription Database on gendatabase.com: “1928 · CORNELIUS ABRAHAM LUUS × SUSANNA FREDERIK ALETTHA BOOYZEN”. The negative was TRUE OF NAAIRS, which is the only index this archive had asked, and it was published as though it were true of the record. A MARRIAGE REGISTER IS NOT AN ESTATE INDEX. A sentence that says “returns nothing” has to name what was asked, and this one did not. Whether this Susanna is the “Ousie” of the Silverton pillar is a separate question and is not being assumed.
how A search run over part of the ground and written up as though it covered all of it.
was “Leeuwberg pinned at FAURESMITH, on a 1903 land file naming a “LEEUWBERG NO 465, DIST. FAURISMITH””
now IT IS A DIFFERENT FARM OF THE SAME NAME. The Friend of the Free State of 13 January 1870 advertises “the farm, WILDEBEESTKOOI, ADJOINING THE FARM LEEUBERG, THE PROPERTY OF THE LATE J.G.E. KOLBE, ABOUT TWO HOURS RIDE FROM THE TOWN OF BLOEMFONTEIN”. Fifteen or twenty kilometres, not a hundred and thirty. The pin stood for two days, and it was published with its own objection printed beside it — that Fauresmith is 130 km from Kaalspruit while the obituary says he died at one and lies at the other. THE OBJECTION WAS THE RIGHT ONE AND IT HAS ANSWERED AGAINST THE PIN. Moved to Bloemfontein, still approximate. The two-farm contradiction this archive could not reconcile was never a contradiction: Leeuberg and Kaalspruit are neighbours.
how A document read wrongly, or a conclusion drawn past what the document says.
was ““1898 — JGL Booyzen of Kwaggafontein, district Marico, IN THE FIELD-CORNET'S PAPERS””
now THE PAPERS ARE THE EDUCATION DEPARTMENT'S. The atlas has carried that line under Marico since the entry was built. The three documents behind it are TAB source OD — the school files of the South African Republic — and they are a subsidy application, book orders and quarterly returns for a farm school. A field-cornet does appear in the bundle, exactly once, valuing the teacher's dwelling in May 1898, and that is a different document sent by a different man. The line was written from a memory of the record rather than from the record. CORRECTED IN PLACE, and the record is now eight documents instead of three.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Margaret Downing “c. 1804, London” — and, for a day, a death date thrown into doubt by a newspaper”
now THE STONE HAD BOTH ANSWERS AND THIS ARCHIVE HAD THE STONE. The Burgersdorp monument has been on the Kolbe page for weeks, photographed and captioned; the caption quoted it as far as “ALSO HIS RELICT MARGARET DOWNING…” AND STOPPED. The next two lines read “BORN IN LONDON 23 FEBRUARY 1803 AND DIED AT BURGHERSDORP 2_ NOVEMBER 1867.” Her birth moves from a guessed year to a date. AND THE DEATH WAS NEVER IN DOUBT: yesterday's Grahamstown Journal obituary said she died “about three months since” — January 1868 — and this archive published the discrepancy rather than resolving it. It was nearly five months, and a journalist rounding is a journalist rounding. THE FAULT IS THE CAPTION: a transcription that stops before the end of an inscription is not a transcription, and this one hid two facts in plain sight for weeks.
how A document read wrongly, or a conclusion drawn past what the document says.
was ““148 rows read of 184 — THIRTY-SIX DOCUMENTS WERE NOT REACHED, and this archive is not going to call 148 a complete enumeration””
now The caution was right in spirit and wrong in fact, and the fact is worse. ALL TEN PAGES HAD BEEN WALKED. The thirty-six were not unreached — THEY WERE THROWN AWAY BY THIS ARCHIVE'S OWN ROW FILTER, which required a description field longer than three characters and so silently dropped every row whose description column the index leaves short or blank. Re-run without that condition the same query returns 194 rows, 184 of them unique: THE EXACT COUNT. This archive published a scope caveat that blamed the source for a hole it had dug itself — which is the same failure as the blank Name column, one layer up: A FILTER THAT DROPS WHAT IT DOES NOT RECOGNISE LOOKS EXACTLY LIKE AN ARCHIVE THAT DOES NOT HOLD IT. The thirty-six turn out to include SEVEN ANGLO-BOER WAR COMPENSATION CLAIMS BY BOOYZENS IN THE MARICO DISTRICT.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Martha Angeline Kolbe, sixty days old, was “BURIED AT VOLKSRUST”, and Volksrust held a concentration camp”
now Both halves withdrawn, six hours after publishing. HER STONE IS IN WAKKERSTROOM MAIN CEMETERY. The index this archive read gives the burial as “Volksrust, Eastvaal District Council, Mpumalanga” — and that is a DISTRICT MUNICIPALITY, not a town; Wakkerstroom lies inside it. This archive read a modern administrative area as a place of burial, which is a failure of reading and not of the source. WITH THE TOWN GOES THE CAMP. The stone reads “TER GEDACHTENIS VAN ONS GELIEFDE KLEINE”, in Dutch, and the mason's town is cut across its base: NEWCASTLE, twenty-five miles away in Natal. A family stone in a town churchyard, cut by a monumental mason in another colony, is not a camp burial. The care taken in the original entry — “records the burial place without claiming the camp” — was not enough, because naming the camp at all in an entry about a child who died in December 1901 makes the claim whatever the sentence says.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The 1902 prisoner of war could not be assigned to either George Augustus Kolbe of Wakkerstroom”
now Now it can, and the withdrawal of the morning is replaced by an answer rather than a shrug. BOTH MEN ARE BURIED IN THE SAME CEMETERY and both stones are read. The younger: “GEORGE AUGUSTUS — GEB. 29 JULIE 1863 — OORL. 28 SEPT. 1954”, which is the Colesberg font date to the day. The elder: “GEORGE AUGUSTUS KOLBE — GEB. 2 OKTOBER 1861 — OORL. 8 NOVEMBER 1934”. THE POW RECORD SAYS AGE 40, EVENT 1902. A man born 2 October 1861 was forty from October 1901 to October 1902; a man born 29 July 1863 was thirty-nine for the whole of 1902. THE PRISONER IS THE ELDER MAN. This is recorded as a correction rather than quietly fixed, because the archive published a claim, withdrew it, and then reversed the withdrawal, all in one day, and a reader is owed the whole of that.
how A document read wrongly, or a conclusion drawn past what the document says.
was “George Augustus Kolbe of Wakkerstroom, born Colesberg 1863, was a prisoner of war in 1902 — “the fourth man of this family taken in that war””
now Withdrawn within the hour, and by the same record that produced it. The POW entry — WO 108/368, the master list of Boer prisoners numbers 1 to 32561 — gives RESIDENCE WAKKERSTROOM, which is what the claim was built on, and this archive stopped reading there. It also gives AGE 40, NATIONALITY TRANSVAALER, PRISON NATAL. The man baptised at Colesberg on 11 October 1863 was THIRTY-NINE in 1902. A SECOND GEORGE AUGUSTUS KOLBE, born at Burgersdorp in 1861 and dead at Standerton in 1934, was forty or forty-one that year — and his own marriage entry to Susanna Margaretha Swart gives his residence as Wakkerstroom too. Two men of one name in one town, which this archive had itself written down as a warning in the same hour, in the same run of searches, and then walked straight past. THE PRISONER IS MORE PROBABLY THE OLDER MAN. Nothing decides it but the age, and an index age is a weak thing; so the claim is withdrawn rather than reassigned.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The Wonderboom deeds register showed a transfer from M. J. Booyzen, widow, to David Schalk Booyzen on 9 April 1907”
now The transferor is right and the date is not. Her gravestone at Rebecca Street, Pretoria, reads “MAGDALENA JOHANNA BOOYZEN (GEB. BEZUIDENHOUT), GEB. 8 MEI 1870, OORL. 29 JAN 1955” — so the ledger's “8/5/1870” beside her name is HER BIRTH DATE, which identifies her beyond doubt. But her husband PETRUS JACOBUS BOOYZEN died on 25 MARCH 1947, so she was not a widow in 1907 and cannot have transferred anything as one that year. The 9 April 1907 read off that row belongs to a different row: the ledger's columns were mis-aligned in the reading. What survives is the identification; what is withdrawn is the date, and with it the suggestion that David Schalk Booysen could be carried to 1907.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Both testable POW ages run one to two years high against this archive's dates”
now That described the De List, not the record. The fuller register gives Stefanus Francois as 26 — exact against a birth of about October 1874 — and David Schalk as 53, within weeks of his fifty-third birthday on 9 June 1848. The De List's 28 and 54 are the outliers. The caveat was published on the strength of the weaker of two documents before the stronger had been found.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The De List's prisoner numbers are a separate series and do not locate these men in the fuller register”
now They are the same series and they locate them exactly. The claim came from misreading a leading digit: image 139 of item 1 begins at 19431, not 18431, which the batch title on image 136 states outright — “Nos. 19,431 to 20,705”. The de Beer entries read as 18463–18465 were 19463–19465. In the correct batch, image 126, numbers 18463, 18464 and 18465 are STEFANUS FRANCOIS, PETRUS JACOBUS and DAVID SCHALK BOOYSEN.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Catherine Maria Sophia is a Barry name”
now It is a KOLBE name. The 1848 Graaff-Reinet baptism register, entry 603, names the godmother in full as CATHARINE SOPHIA MARIA KOLBER — eleven years before her sister named the daughter this archive had been treating as the head of the chain. The namesakes page called it “a Barry name that outlives the Barry surname by eighty years”; it outlives the Kolbe surname instead, and the archive already held the woman, filed as plain “Catharine Kolbe”, without noticing she carried the name first.
how A document read wrongly, or a conclusion drawn past what the document says.
was “A relationship search had found one name the Barry notices never gave”
now It found none. HENRY JAONAS BARRY is JAMES HENRY BARRY, who was already in this archive — the same baptism indexed twice, once as Henry Jaonas with the year read as 1868 and once as HENNERY JAMES BERRY with it read as 1848. Same birth date, same parents, and FamilySearch attaches both to one tree person. The register itself decides: “Graaff-Reinet. Baptism Records 1846–1851” cannot hold an 1868 entry. So the “baptised at twenty years old” that made him interesting was a mis-transcribed year, and “Jaonas” was a mis-read “James” — which is what the archive's own register search implied when it found JONAS nowhere among the Barrys or Kolbes. The eleven-against-nine gap is unchanged: TWO children are still missing, not one.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Magdelena Elisabeth de Beer was carried here under WikiTree's codes, b6c2 and b6c6”
now The codes are b5c2 and b5c6. de Villiers 1893 page 34 puts ZACCHARIAS DE BEER — her father — as the FIFTH child of Matthijs Andries de Beer, not the sixth, and Pieter van Zyl said so on G2G before the book was opened; reading page 34 confirmed it. The b6 is WikiTree's, and this archive repeated it for a day without checking it in print. It matters more than a digit: a de Villiers code is an address, and quoting the wrong one sends the next reader to the wrong family. Where this archive QUOTES the WikiTree profile it still says b6, because that is what the profile says — the error being reported is not the same as the error being made.
how A claim taken from the family tree and repeated before it was tested.
was “The two Magdalena de Beers were two different women, and the tree had merged them”
now They are one woman. de Villiers 1893, page 34, marks ONLY the second — Magdalena baptised 16 October 1754 — with “zie Booijsen”, and 1754 is the birth year this archive has always carried for its own Magdelena Elisabeth. The argument published here for a day ran the other way: that Jan Booysen's wife was dead by December 1793 while ours died in 1843, so they could not be the same. The 1843 death date is the weak link, not the identity — this archive's own page said both her dates came off the family tree with no source stated. What actually followed from reading the book is the opposite of what was published: she married JAN, and generation seven is now in question rather than confirmed.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “A relationship search had turned up two Barry children on neither parent's death notice”
now It turned up one. WILLIAM FREDERICK THOMAS BARRY, baptised 29 September 1861, was already recorded on William Barry's own page — put there the day the Burgersdorp doopregister was read, with the doubt stated in the same sentence. The search re-surfaced him and this archive wrote him up as a find for about an hour. Only HENRY JAONAS BARRY is new. The section now says which is which, and the header no longer counts him.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Hendrik Stephanus Nicolaas Booyen was buried at Johannesburg”
now He was buried at ROODEPOORT–MARAISBURG, twenty kilometres west, in the Central Cemetery there — section I D.R., entry 96. The FamilySearch collection is titled “Johannesburg Cemetery Records” and the index repeats that label for every register inside it, including those of other municipalities. The collection title is not the place.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Anna J. Mary Booyzen and George Downing Daschner were stillborn or died the same day”
now Both lived. She was SIX MONTHS old and he was THREE MONTHS AND FIVE DAYS. The Brixton register keeps Years, Months and Days in three separate columns and FamilySearch's index collapses them into one field, so a child with the Years column blank comes out as “0 days”. This archive printed the index reading for an afternoon without opening the page.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Only a family tree puts Anna Maria van der Merwe in this family”
now A Dutch Reformed baptism does. Her son DAVID SCHALK BOOIJZEN was born 12 December 1806 and christened on 4 January 1807, with PIETER JACOBUS BOOIJZEN and ANNA MARIA VAN DER MERWE named as his parents — and he is first of the eight on the 1876 Sterfkennis. The wall on the ancestor chart said the wrong thing for a day and has been rewritten.
how A search run over part of the ground and written up as though it covered all of it.
was “No Booyzen filed a war-losses compensation claim”
now One did. This archive searched the War Losses Compensation Commission, found nothing, and published the zero while noting it was weak because the Commission's records are catalogued by sub-commissioner and not by claimant. David Schalk Booijzen's answer to question ten of his wife's 1900 death notice is “none. Compensation claim filed.” The claim exists; it was the finding aid that could not reach it.
how A negative built from a list mistaken for a population — the archive searched what it happened to have and reported the answer as though it had searched everything.
was “Anna Catharina Johanna Olivier was on this site twice”
now One woman, two records: a stub written from the 1908 death notice — “His first wife, named as pre-deceased” — and a full record written from her own 1891 Sterfkennis hours later. The shared-name page gathered them as though they were two people. Merged. The build now warns on every shared name so a human checks each one reads as two lives and not one.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Charlotta Elizabetha Dormehl died before 1908”
now She died on 6 February 1902, at Driefontein in Groot Marico, and her own Death Notice 844 says so. It also gives her father as Jacob Dormehl, her mother as unknown, her age as 60 years 4 months, her estate as one wagon and her furniture, and THREE husbands where this archive had one — Jacob van Staden, Matthys Roos, then Booysen.
how A search run over part of the ground and written up as though it covered all of it.
was “Anna Johanna Booijzen, born Alberts, died at Boekfontein and is not this family”
now Every clause of that was wrong. She is Anna Catharina Johanna OLIVIER, she died at BRAKFONTEIN — the family's own farm, where the archive's oldest document was written — and she is the first wife of Hendrik Stefanus Booijsen. The maiden name reads Oliefier and contains no b, r, t or s; her son gives OLIVIER on the 1908 notice. Two misreadings on one page kept a documented wife, seven children and a whole Olivier line off this site for a week.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Hendrik Stefanus Booysen had four children”
now He had at least seven. His first wife's 1891 Sterfkennis lists them all, of age: Petrus Jacobus, Ocken Johannes, Anna Johanna Maria, Elesabettha Maria Cornelya, Alettha Catrina Johanna, Magdalena Johanna and Hendrick Stefanus Nicolaas. The 1908 notice names four because by 1908 only four were left.
how A search run over part of the ground and written up as though it covered all of it.
was “Hendrik Stefanus Booysen's four children were his second wife's”
now They were his first wife's. The Zeerust marriage register, entry 13, dates the second marriage to 14 April 1892 and gives the bride's age as fifty-one and her condition as widow. By 1908 all four children were adults, two of them married women. The archive put them under the marriage it could name rather than the marriage that could have produced them.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Margarietha Wilhelmina Elizabeth Barry was John Augustus Barry I’s daughter”
now She was his granddaughter. The 1915 death notice names eight children and she is not among them; her father is Hendrik Nicholaas Barry, whose other children were baptised at Dordrecht and Slang River between 1888 and 1900. Moved down a generation — which is where the cousin marriage of 1924 needs her to be, and the archive had her in the wrong place while asserting it.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The 1915 notice named eight children and the archive listed nine”
now Two men on that list — John Augustus Barry II and George Augustus Barry — had no page at all, while a ninth name stood in their place. All three fixed together; the build now refuses to publish a household naming somebody who does not exist.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “The ancestor chart, every wife one generation too high”
now The chart put each man's wife in his mother's box. Kathleen Barry stood as Willem Hermanus Booyzen's mother when she was his wife; Johanna Catharina Mountjoy stood as Petrus Jacobus Booysen's mother when she was his; and the woman at the root had no mother drawn at all. Four slots, all shifted by one. Rebuilt, and the genuinely empty slot — the mother of the 1812 man — is now a wall.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Two Barry sisters married Booysens”
now One did. The 1915 death notice lists the siblings and there is no second.
how A claim taken from the family tree and repeated before it was tested.
was “The farm was called Armadale”
now No document anywhere names a family farm. The name came off the tree and had nothing behind it.
how A claim taken from the family tree and repeated before it was tested.
was “Orphaned at seventeen”
now Eighteen. Her father died in December 1947 and she was born in March 1929.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Anna Pretorius born 1905”
now 1906. A baptism register and her own death form agree against the tree, and the tree had mistranscribed one digit.
how A claim taken from the family tree and repeated before it was tested.
was “The Farm List returns nothing, so the farm was not burned”
now The negative was VOID. The register searched turns out to be an Orange Free State land roll that does not cover the Cape at all.
how A search run over part of the ground and written up as though it covered all of it.
was “The tree contradicts itself on 1827 and 1840”
now It does not. 1827 is when James Montjoy died and 1840 is when the notice was filed. The contradiction was the archive's.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The z was this family's invention”
now The 1851 Graaff-Reinet register writes a neighbouring family's name with a z in the same hand and ink. The z was Cape Dutch orthography before it was anybody's statement.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Willem Hermanus is a Booyzen name”
now It is a Pieterzen name. It came in through a step-relation with no blood tie to this family at all, and four generations have carried it.
how A document read wrongly, or a conclusion drawn past what the document says.
was “A distribution chart with nine provinces”
now Five of the nine had been filled in by the archive itself. The source reports four. The chart now draws four, and a dash where nothing was reported.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “The family tree renders correctly”
now It did not. Auto-linking injected HTML anchors into an SVG and silently erased half the names in it. It did not look broken; it looked plausible.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “MOUNTJOY returned 42 documents, all read”
now All twenty-two dated 1930 or later were read. The twenty before 1930 were not, and they held James Mountjoy's wills of 1818 and 1827 and an 1834 Colonial Office memorial. A filter applied for convenience was written up as the whole search.
how A search run over part of the ground and written up as though it covered all of it.
was “The SANDF enquiry was sent”
now It was sent, and one of the two addresses bounced at the server. Sent and delivered are not the same word and this archive used the wrong one.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “The 1905 clerk's Booysen proves the family used an s”
now It proves nothing. Seven Attorney-General files spell one Middelburg constable both ways, twice inside 1905. Only the son's signature still carries weight.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The archives hold 39 Booyzen documents”
now They hold 184. Every search in the project had been run against the Cape depot alone, one repository out of nine, for weeks.
how A search run over part of the ground and written up as though it covered all of it.
was “Maggie Booyzen was born 1 October 1926”
now 14 October. Her father wrote it on her mother's death notice three months before he was widowed.
how A claim taken from the family tree and repeated before it was tested.
was “1944 is the one time the form worked”
now The very next document opened named both parents too. The claim was a count of documents read, dressed as a count of documents that exist.
how A negative built from a list mistaken for a population — the archive searched what it happened to have and reported the answer as though it had searched everything.
was “The Marico Booyzens are not this family”
now The direct ancestor died in district Rustenburg in 1876, twenty-five years before the war. The reason for holding the two clusters apart has gone; the atlas now says unresolved.
how A negative built from a list mistaken for a population — the archive searched what it happened to have and reported the answer as though it had searched everything.
was “Not one miner among them”
now There is one. A son was a MIJNWERKER of Dugmore Mijn at Indwe in 1908. The negative was built out of six occupations the archive happened to have read, and it had not read his because it did not know the man existed.
how A negative built from a list mistaken for a population — the archive searched what it happened to have and reported the answer as though it had searched everything.
was “He died in the Transvaal at about thirty”
now Estate O/2899 belongs to his uncle, born 1818. The death date had no source but the file number: the archive read a year off a reference, wrote it in as a fact, and cited the fact back.
how A fact derived from a reference, written into the record, and then cited back as evidence.
was “Isabella Frederika Coetsee died in 1943”
now 31 March 1930. 1943 is the year her husband finally filed — thirteen years late. The same mistake as O/2899, made in the same way, and found the same afternoon by opening the document instead of trusting the reference.
how A fact derived from a reference, written into the record, and then cited back as evidence.
was “John Augustus Barry was born at sea”
now Born in IRELAND. The 1903 registration of his death, made by the son who was present, enters him as “Irishman born in Ireland”. The 1915 notice this archive had been using was sworn by another son twelve years later and says “at sea, exact whereabouts unknown”. The archive had the weaker document and had not looked for a better one.
how A search run over part of the ground and written up as though it covered all of it.
was “He died on 3 August 1903”
now 25 August 1903, as the family tree said all along. The 1915 notice gives the 3rd; the statutory registration, signed by a son present at the death, gives the 25th. For once the tree was right and the document this archive preferred was wrong.
how A search run over part of the ground and written up as though it covered all of it.
was “Dorothea Regina Barry was a daughter of John Barry”
now She is not on the 1915 death notice. This archive said she was “named on” it, having quoted four lines from the top of the page without reading the children's list at the bottom. The only Dorothea Regina in the registers is a granddaughter, baptised at Slang River in 1892.
how A search run over part of the ground and written up as though it covered all of it.
was “Ruben Jacobus Josuah Barry was a son of John Barry”
now Also not on the notice. A Ruben Jacobus Hosea was baptised at Dordrecht in 1888 — a grandson. The Schoeman name Ruben does come into this family, one generation later than this archive said.
how A search run over part of the ground and written up as though it covered all of it.
was “Johan Gottlieb Kolbe was buried at the German church of the Savoy, entry 55, grave 7”
now Taken from a family tree and repeated here. The Westminster Burials register, held by the City of Westminster Archives Centre, has him DIED 18 NOVEMBER 1845 AND BURIED 26 NOVEMBER 1845 AT ST JOHN THE BAPTIST, GREAT MARLBOROUGH STREET, page 7. The date matches to the day and the seven matches; the church does not. The register also gives a death date and a birth year of 1764, neither of which this archive had.
how A claim taken from the family tree and repeated before it was tested.
was “Johan Gottlieb Kolbe had one wife, called Mary”
now He had two, and both were called Mary — which is why one of them could never be made to fit a run of baptisms from 1792 to 1819. The first died 2 August 1814 aged 46; the second died 15 August 1824 aged 36. Both are in the German Lutheran Church register, TNA RG 4/4631.
how A document read wrongly, or a conclusion drawn past what the document says.
was “R_177176049779 is this archive's reference for Petrus Jacobus Booysen's 1812 birth”
now It was recorded once, from a FindMyPast search run without a subscription, and never opened. Opened twice on a live full subscription, FindMyPast's own transcript page returns a 500 error. THE REFERENCE DOES NOT RESOLVE. It stays on the page as a struck citation rather than being quietly deleted.
how A fact derived from a reference, written into the record, and then cited back as evidence.
was “The German Savoy registers are RG 4/4625, 4626, 4628 and 4631 for burials”
now RG 4/4630 belongs on that list too. The 1807 burial of Charles Samuel Kolbe is entered in both 4631 and 4630, with the same death date and the same age to the day — a register and its duplicate. The archive's own note named five volumes and missed one.
how A search run over part of the ground and written up as though it covered all of it.
was “Seventeen people had a page saying they are “not yet discussed on any page””
now Among them GEORGE AUGUSTUS KOLBE, the 1820 Settler, named 155 times in the built site; JOHN BARRY, 158; GEORGE MOUNTJOY, 140; PETRUS JACOBUS BOOYSEN, 110; JOHANNA CATHARINA MOUNTJOY, 89. THE DATA WAS FINE AND THE BUILD THREW IT AWAY. dossiers.py dropped any name that was a strict prefix of a longer person’s name — a rule written to stop “John Barry” swallowing “John Barry Booyzen”, which instead silenced the shorter man entirely. The fix keeps the shorter name and fences it with a negative lookahead, so it matches everywhere except where the longer name begins. 2,428 mentions came back. NOTHING REFUSED THE BUILD WHILE THIS WAS TRUE, so a gate now does: checkregisters.py fails when a person the dossier calls undiscussed is named in the built pages. Found because the estate asked this archive to look for a fault the Mazza archive had just had.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “WO 12/1430 is the muster book to read for James Mountjoy”
now It is the 20TH AND 21ST DRAGOONS FOR 1867–1868 — fifty years after he was discharged, and the re-raised regiment rather than the one that garrisoned the Cape. The 21st Light Dragoons has a continuous WO 12 run of its own and the piece that covers his discharge is WO 12/1450, 1816–1817. THE WRONG REFERENCE HAD BEEN WRITTEN INTO A LETTER TO KEW; the letter was still unsent and now names the right piece and says an earlier version had it wrong.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The 21st Light Dragoons left the Cape in 1816”
now September 1817. An extract of a letter from the Cape dated 1 and 10 September 1817, printed in the Sun, the Star, the Public Ledger, the Morning Herald and the Hampshire Chronicle that November: “the 21st Light Dragoons have embarked for India”. The 1816 figure was written on 15 September with no source attached and stood in three files. IT MATTERS BECAUSE HE MARRIED IN MARCH 1817 — under 1816 he was already a civilian when he married, and under 1817 he had six months to choose between a wife at Graaff-Reinet and a transport to Madras.
how A search run over part of the ground and written up as though it covered all of it.
was “The Burgersdorp sweep of images 605–818 found five new people”
now Three. Johanna Catharina Mountjoy, born 28 January 1860, was already held in full — a daughter of George Mountjoy who married William Henry Keys — and the sweep only confirmed her and gave her a reference. Frederick Fortunatus Werdmuller was already held from a 1907 petition and a 1953 estate; what the sweep added was his birth and his mother, which is a real gain and not a new person. THE DUPLICATE PERSON RECORDS WERE CAUGHT BY THE BUILD'S OWN NAME-COLLISION CHECK for the first and by reading the register for the second, before either reached a page. The rule it costs: a sweep that returns a name is checked against the archive before it is called a find.
how A search run over part of the ground and written up as though it covered all of it.
was “MSB 751 is a letter written to George Augustus Kolbe “on his ordination””
now It is not. The National Library’s catalogue described it that way and this archive repeated the catalogue without ever asking whether the date fitted. The DSAB draft that came with the letter dates the ordination to 25 MAY 1833 — two years and nine months after the letter of 23 August 1830 — and the letter itself says George had been “fully accepted by the [L.M.S.] as one of their Missionaries”, which is a different step. THERE WERE THREE STEPS AND THE ARCHIVE HAD COLLAPSED THEM INTO ONE: mission assistant on 8 February 1828, accepted as a missionary in the summer of 1830, ordained by Arbousset and Casalis on 25 May 1833. A CATALOGUE DESCRIPTION IS A FINDING AID, NOT THE DOCUMENT. This archive quoted one as though it were the second, and the document itself corrected it the day it arrived.
how A fact derived from a reference, written into the record, and then cited back as evidence.
was “The Settler’s parentage — father AND mother — is on the parish register”
now The father is. The mother is not. The Westminster baptism of 2 February 1803 names “John Gottlob and Mary”, and “Mary” is all it gives, at that font and at eight others in the same family. The German Lutheran burial register of 1814 gives her death, her burial and her age of 46 and NO MAIDEN NAME. “MARY LARMAN” COMES FROM THE FAMILY TREE AND FROM NOWHERE ELSE THIS ARCHIVE HAS READ — and this archive’s own person record has said so all along (“her surname before marriage is not known and no marriage has been found”), while the page beside it put the tree’s surname inside a sentence ending “is on the parish register” and chipped it Documented. THE DATA WAS RIGHT AND THE PROSE WAS WRONG. Found while testing a FamilySearch memory of 2017 that makes the same claim; the search engine returned this archive’s own repository as the authority for it, WHICH IS WHAT CIRCULARITY LOOKS LIKE FROM THE INSIDE.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The DSAB typescript is unsigned, and this archive will not name its author”
now IT IS UNSIGNED. THE EMAIL THAT CARRIED IT IS NOT. Melanie Geustyn of the National Library, with the Background File open in front of her, wrote: “I am also sharing with you a two page document that was written for the DSAB … The author, KS, WOULD BE DR. KAREL SCHOEMAN who worked here at the National Library.” This archive opened the two attachments, read four pages of manuscript and two of typescript, reasoned its way to Schoeman from the inside — the Bethanie claim has only his 1985 book behind it, and the paper was typed where he worked — and published the attribution as an inference it would not assert. THE ANSWER WAS IN THE COVERING SENTENCE, UNREAD. The Background File also holds nothing further about the donation or about any other deposit by Dr Griffiths, which answers a question the enquiry asked and which this archive likewise did not notice it had been given. THE RULE: READ THE COVERING LETTER, NOT ONLY THE ENCLOSURE. An archivist who goes to the trouble of answering in prose is a source, and the prose is part of the reply.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The list of what MSB 751 contains — published as “WHAT IT DOES CARRY””
now IT WAS A SAMPLE PRESENTED AS AN INVENTORY. The letter was first read in a mail client’s PREVIEW OVERLAY on a screen eight hundred pixels wide, because this archive had declined to download the files. Nothing published from that reading was wrong. Opened properly as a PDF two days later, the same four pages also carry: DR PHILIP IN LONDON, “exceedingly glad” at the acceptance and promising to write to George himself — where this archive had him only as a courier; A JACKET AND WAISTCOAT FOR “JOHN” with “tell him his old father has sent it him, & to make haste to fill them out”; A YEAR OF NEWSPAPERS, one number lost by Henry “who has them every week, after I have done with them”; and FIVE NAMED FRIENDS OF THE HOUSE — Mr and Mrs Stanhope, Mr and Mrs Warner, Mrs Field and her daughters, and Miss Mary Field who had married a Mr Bellew, a surgeon. THE RULE: A DOCUMENT READ IN A PREVIEW IS NOT A DOCUMENT READ. If it is worth quoting it is worth opening.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The DSAB entry is wrong about exactly one thing, and its first source is a reference this archive did not have”
now BOTH HALVES WRONG, AND BY THE SAME CAUSE. The eleven-item source list was worked through WITHOUT FIRST BEING CHECKED AGAINST THIS ARCHIVE. Source 1, the death notice MOOC 6/9/35 no. 7698, HAS BEEN READ HERE SINCE BEFORE SEPTEMBER — it is published with two photographs and a transcription — and a work-list row was raised to order it. Struck. AND BECAUSE THE NOTICE WAS ALREADY IN HAND, THE ENTRY’S SECOND ERROR WAS THERE TO BE SEEN AND WAS NOT: Schoeman writes that the death notice records only eight children. IT NAMES TEN. So the archive published “wrong about exactly one thing” while holding the document that shows a second. The same reading also over-claimed Knapdaar as a new name to search; this archive already had it, from that same notice. THE RULE, WHICH IT ALREADY HAD AND DID NOT APPLY: a sweep that returns a name is checked against the archive before it is called a find — AND A BIBLIOGRAPHY IS A SWEEP.
how A search run over part of the ground and written up as though it covered all of it.
was “The Pretoria estate files are wanted “for the parents” — and a researcher was about to be paid on that reason”
now THE PARENTS ARE ALREADY DOCUMENTED, FROM CHURCH REGISTERS, AND HAVE BEEN FOR DAYS. Catharina Maria Sophia Booysen was baptised at BARKLY OOST, 24 MAY 1891, ENTRY 120; her brother Henry James at BARKLY EAST in 1887, the register naming WILLEM HERMANUS BOOÿSEN AND CATHARINA MARIA SOPHIA BARRY in terms. The brief of 15 September asked for six death notices “for the parents, the birthplace, the occupation and the list of children”, and that brief was still being repeated on 20 September in a letter about money — MHG 3734/46 described as “the one I most need, for her parents”. DAVID CAUGHT IT, from a Google result, by asking what the files were actually for. HIS SEARCH RETURNED MYHERITAGE, WHICH IS THIS FAMILY’S OWN TREE READ BACK TO IT — circular, and not evidence — BUT THE CONCLUSION WAS RIGHT ANYWAY, because the archive holds the registers behind it. THE FAULT IS THE DAY’S THIRD OF ONE KIND: a list worked without first being checked against what is already held. The files have been re-justified one by one on what is genuinely missing, WHICH IS CHILDREN AND NOT PARENTS.
how A search run over part of the ground and written up as though it covered all of it.
was “FindMyPast barely covers London baptisms for 1800–1806, so the missing Downing baptism proves nothing”
now THE COVERAGE IS EXCELLENT AND THE NIL IS REAL. Published within the hour on the strength of a count: 1,131 Downing baptisms in England for 1800–1806 and almost none in London. THAT WAS THE WRONG CONTROL — counting one surname by county tests where the Downings were, not whether the index reaches London. THE RIGHT CONTROL IS A COMMON NAME: of 100 SMITH baptisms sampled in the same window, THIRTY-SEVEN ARE LONDON, MIDDLESEX OR WESTMINSTER; of 76 KOLBE baptisms, six. The index covers London well. So the near-absence of London Downings is not the index failing — THERE WERE VERY FEW DOWNINGS IN LONDON. AND THE CORRECTION REVERSES THE ARGUMENT TWICE: the nil is a real negative, and a family with a London birth in 1803, no London baptism and no London kin in the registers IS WHAT A FAMILY NEWLY ARRIVED FROM THE COUNTRY LOOKS LIKE — which strengthens the Welcombe marriage of 1802 instead of weakening it. THE RULE THIS ARCHIVE ALREADY HAD AND MISAPPLIED: a nil return is worthless until a control shows the search works, AND THE CONTROL HAS TO TEST THE THING YOU ARE DOUBTING.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “The Welcombe marriage of 10 May 1802 is a candidate for Margaret Downing’s parents, and a strengthened one”
now IT IS NOT. WITHDRAWN THE SAME DAY IT WAS RAISED. Published this morning as a candidate on a name match and a nine-month interval; called STRENGTHENED at lunchtime when a London baptism could not be found for her. THE QUESTION NOBODY ASKED UNTIL EVENING WAS WHETHER THE WELCOMBE COUPLE STAYED. They did. Every Downing record in that parish reads: 1802 JOHN MARRIAGE, 1803 JOHN BAPTISM, 1806 William baptism, 1809 Emmey baptism, 1810 Elizabeth baptism. THEY BAPTISED A CHILD AT WELCOMBE THE YEAR AFTER THE WEDDING AND WENT ON DOING IT. Margaret was born in LONDON in February 1803. AND THE LAST LINE OF THE PARISH LIST EXPLAINS THE WHOLE THREAD: a MARY DOWNING BURIED AT WELCOMBE IN 1834, which is the year both trees give Margaret’s mother. A previous researcher almost certainly found the one Downing–Pooley marriage in England and a Mary Downing burial in the year they wanted, in the same tiny parish, and joined them — the same search this archive ran today, stopping one question earlier. SO “MARY KING POOLEY, 1768–1834” IS PROBABLY NOT A RECORD BUT A RECONSTRUCTION, which also explains why this archive’s own copy of the tree calls her MARY P KING: two compilers, the same fragments, two guesses.
how A document read wrongly, or a conclusion drawn past what the document says.
was “www.familysearch.org is down, and so are The National Archives at Kew and the University of Edinburgh”
now NONE OF THEM. /sources/ published that table for five days, and FAMILYSEARCH IS CITED 7,276 TIMES ON THIS SITE. FamilySearch answers 403 — a bot check — and Kew and Edinburgh answer 200; all three were used from this desk on the day the table was found to be wrong. THE CAUSE: checkexternal.py identifies itself honestly as “TheBooyzenArchive”, and firewalls refuse unfamiliar user-agents at the connection, before any HTTP status exists — so the tool’s own sensible rule that 403 means GUARDED never ran, and a refused connection was written down as DOWN. THE FIX IS NOT TO DRESS UP AS A BROWSER. This archive identifies itself and will go on doing so. The fix is to stop calling a silence a death: there is now a state UNVERIFIED, kept apart from DARK, which means does not resolve in DNS and is the only state here that means gone. AFTER THE FIX: 27 up, 11 unverified, 5 mail-only, 4 dark, and NO HOST CLAIMED DOWN. The rule was on /method/ the whole time — a nil return is worthless without a control, and a refused connection is a nil return with no control.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “Commit d3828d3, “The tailor was right: it is James Hogg”, is a record of that afternoon’s work”
now IT IS A RECORD OF TWO SESSIONS’ WORK UNDER ONE SESSION’S MESSAGE. It was staged with git add -A scoped to three directories — which is still -A — while the landing-site session had staged surgically and was waiting on a build. ALSO IN THAT COMMIT, AND NOT ITS AUTHOR’S: GENERAL WILLEM JOHANNES KOLBE (1847–1941), who led the Bloemfontein commando, was made general at the siege of Kimberley and rode in De Wet’s Cape invasion — and whom this archive already held as one line in a prisoner list; the Herinneringe van genl. W J Kolbe memoir at the Free State Archives; the finding that the DSAB was terminated after Volume V in 1987, so Schoeman’s February 1992 draft was written for a series that had already stopped; the discovery that two sections of /documents/ shared the id pow-1901 with fourteen links reaching the wrong one; and the #pow-wo108 rename that fixed it. THE COMMIT IS NOT AMENDED. It is pushed, two sessions share this tree, and FORCE-PUSHING TO REWRITE SHARED HISTORY WOULD BE A WORSE FAULT THAN THE ONE IT CORRECTED. This archive records its mistakes rather than erasing them, and row 160 had already logged this hazard four times — every one of them pointing the other way.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Every person page printed a mention count — “N mentions across N pages” — as a count of where that person appears in this archive”
now THE NUMBER WAS ABOUT TWO AND A HALF TIMES THE TRUTH, ON ROUGHLY 350 PAGES, FOR AS LONG AS THE TOOL HAS RUN. It published 24,981 mentions where the real figure is 9,556. tools/dossiers.py counts a person by scraping the BUILT site — and the built site prints what the tool wrote last time. Every person page carries an “Every mention, page by page” section holding the stored excerpts, so 7,307 of 9,008 excerpts, 81%, were harvested from person pages: four fifths of what the tool read was its own previous output, re-rendered. The counts came back too, quoted inside the excerpts — 1,732 of 9,008, 19% — because the pages print the numbers as well. THE FILE NEVER SETTLED. Generate, rebuild and generate again and all 347 dossiers changed, 24,981 down to 24,636, and again the cycle after. Two consecutive runs were byte-identical and proved nothing, because both read the same build. The test that catches it is generate → rebuild → generate → diff, and it now runs as a gate: checkregisters.py refuses a build where a person page prints a different number of excerpts than the file holds. Found by the landing-site session, which asked why the diffs were unreadable and would not accept two identical runs as an answer. The tool now strips the generated section before it reads a page, and the count converges: three cycles in a row at 9,556, unchanged. WHAT A READER LOST AND DID NOT LOSE. The count under a name was wrong and the quotations chosen were partly drawn from the archive quoting itself. No document, date, name or relationship anywhere in this archive was derived from these numbers — nothing is sourced to a mention count, and no page sums them into a total. The two large “N times” figures on /sources/ and /register/ come from elsewhere and are untouched.
how A fact derived from a reference, written into the record, and then cited back as evidence.
was “Commit e40087d says the regenerated dossiers changed order and nothing else — “345 are reordered and nothing else … zero have changed content””
now THE TEST THAT PRODUCED THAT SENTENCE COULD NOT HAVE RETURNED ANY OTHER ANSWER. The comparison did sorted(json.dumps(x) for x in a[k]) where a[k] is the dossier DICT, not its hits list — so it iterated the KEY NAMES, n, slug, mentions, pages, hits, which are identical for every person in the file. The “reordered only” branch was guaranteed true. It was not a weak test; it was a test of nothing, and it returned the reassuring answer because that was the only answer available to it. Redone against hits: identical 2, reordered-only 0, content changed 345, and anna-catharina-sleer shares 12 of its 26 excerpts with the version before. So 345 dossier pages showed a reader different quotations than the day before — a reader-visible change, written up in a commit message as a tidy-up. Caught by the landing-site session, which ran the same comparison order-insensitively and got the opposite result. The commit is pushed and is not being amended; this row is the record. The rule is the one this archive has now logged three times in a day — an untested reference is untested code, a number read off an unstable object is not a number, and a check that cannot fail has not checked anything.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was ““FindMyPast is the wrong tool and that is worth writing down: it indexes Württemberg almost entirely as a birthplace field in emigration records, not as German parish registers. Kolbe + Württemberg returns 0 there against a working control.””
now FINDMYPAST HOLDS A GERMAN MARRIAGE INDEX, AND THE ZERO WAS A BROKEN FILTER. Germany Marriage Index 1558–1929 is a FindMyPast record set holding 338 Kolbe marriages, giving both parties, the year and the parish — and it reaches Württemberg: Agnes Kolbe × Jerg Raisch, 1685, Evangelisch, Groetzingen, Schwarzwaldkreis, Wuerttemberg, and Albrecht Jonathon Kolbe, 1741, Knittlingen, Evangelisch, Neckar, Wrtt. THE PLACE FILTER ON THAT DATASET RETURNS 0 TO EVERYTHING. place=wuerttemberg returns 0 against records read minutes earlier; place=knittlingen returns 0 against a Knittlingen record on the same page; and place=westfalen returns 0 when seven of the eleven Kolbe marriages in the 1778–1798 window are Westphalian. The dataset filter works and the year filter works — 338 cuts to 11 — and only place is dead. So the earlier “working control” was a control of the wrong thing. It showed that FindMyPast answered, not that the place filter applied, and a filter is not tested by a search that does not use it. What survives, and it is narrower and better: no Gottlieb Kolbe marries anywhere in that index between 1778 and 1798, all eleven Kolbe marriages in the window are Prussian, and the index's coverage in those years does not reach Württemberg at all. That is still not evidence he did not marry there. The rule, met twice today: before believing a filter's nil, search it for something you have already seen with your own eyes.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “Work list row 177, on the 2012 Barry register report: “There is no Casmier, Casimir or Kasimir anywhere in this archive.””
now HE IS IN BOTH OF HIS PARENTS' DEATH NOTICES. The Settler's notice, signed M. Kolbe — Margaret, the widow — at Colesberg in January 1845, names ten children and JOHAN CASIMIR KOLBE is the fifth of them. Margaret Downing's own notice of 25 November 1867, signed by her son-in-law G. D. O. Werdmuller, names him again — and that document is careful about the dead, entering Mary Elizabeth as “(dead), who was married to George F. Enslin” and him plainly. So he was alive at about twenty-three. Both transcriptions have been in notes/documents-transcribed.md since 10 September. He is also on the archive's own Kolbe page, in the list of children, with the variant spelling “Casiner” already recorded — and the same page's prose singles out Casimir by name as one of the family's extraordinary forenames. WHAT WENT WRONG IS THE SHAPE OF THE SEARCH. The archive's people list was searched and the archive was reported on. people.js is where a person's record lives; it is not where the archive's evidence lives, and this man had no record precisely because nobody had joined the notice to the list. A search of the index is not a search of the holdings, and the one place he was missing is the one place that was looked. The compiled report was right and this archive was wrong about its own documents — its “Casmier” is a third spelling of a name in two primary sources. He now has a record.
how A search run over part of the ground and written up as though it covered all of it.
was “Commit e212628: a build failure that afternoon “did not reproduce. Green on the re-run — one failure is the weather.””
now IT WAS NOT THE WEATHER. IT WAS A BUG IN build.sh THAT HAD BEEN THERE ALL DAY, AND THE RE-RUN PASSING IS EXACTLY WHAT IT LOOKS LIKE. The two build steps read cd site && npm run build >/dev/null && cd .., and set -e does not abort on a failing command inside an && list. Proven with a three-line script: cd site && sh -c "exit 143" && cd .. falls through to the next line, leaves the working directory in site/, and the script exits 0. So when astro was reaped, the pass failed, cd .. never ran, the script carried on, and python failed to find site/tools/links.py — two steps downstream of the real event, with the trap faithfully reporting “NOT KILLED, exited 2 during: links.py”. The re-run passed because nothing killed it that time, not because there was nothing wrong. AND THE WORSE HALF: A KILLED BUILD COULD HAVE EXITED 0. The three-line proof does exactly that. If the reaped step had been followed by steps that happened to pass, build.sh would have printed “build complete” over a build that never finished. Found only because the D'Arcy session asked this archive to test its own stopper, which meant killing a live build on purpose — the first time a build here was killed deliberately rather than by the machine. The fix is a subshell, ( cd site && … ), which restores the directory by construction and propagates the failure, so set -e fires on the step that actually failed. Verified by killing a build mid-flight: it now stops at “KILLED by signal 15 during: first pass”, with nothing surviving. THE RULE THIS BREAKS IS MY OWN, STATED THE SAME DAY: three identical failures is a fault and one is the weather. That is a decent heuristic for load and a bad one for a bug that only shows when something else goes wrong, because the trigger is intermittent and the fault is not. A failure that does not reproduce has not been explained; it has only been survived.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “TAB MHG 52653 is not a file — the reference is probably wrong, and what is wanted is the correct number from the paper MHG index for the 1920s”
now IT IS A FILE, IT IS AVAILABLE, AND THE RESEARCHER SAID SO ON 15 SEPTEMBER. Anne Brest answered this a week before the archive published it as an open question. She pasted a NAAIRS link to the record itself and wrote: “MHG 0 52653 BOOYZEN, JACOBUS NICHOLS. THE DATE IS 1923. DON’T WORRY that it is not appearing after the 52653 number. . some references are like this.” The missing year suffix is normal for some references — and that missing suffix is the exact feature this archive had treated as evidence the reference was broken. SHE WAS THEN TOLD TWICE NOT TO HUNT IT, and said noted each time, having already found it. The claim above went out in a letter to her on 21 September, and she corrected it the next morning: “THIS IS A FILE … it IS available and I have it on the list. I did explain this in the last email.” THE ARCHIVE’S OWN OBSERVATION WAS NEVER WRONG — it genuinely could not find 52653 on microfilm, and it is not name-indexed under any spelling. What was wrong was the inference: from not on film to the reference is bad, when the answer was that it is a paper reference of a shape the 1920s used. THIS IS THE SECOND TIME IN A FORTNIGHT that a correspondent answered a question and this archive read past the answer — the first was Melanie Geustyn naming the DSAB’s author in her covering sentence. The rule stands and was not followed: READ THE COVERING LETTER, NOT ONLY THE ENCLOSURE. A researcher who tells you a thing is findable has told you something no index can.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Commit 5d5e90e, in this archive’s own history: “Every gate in this archive has been grading somebody else’s build … this archive’s package.json passes --dist dist on the command line.””
now TRUE OF SEVEN ARCHIVES AND NOT OF THIS ONE. This archive's package.json has passed --dist "${ARCHIVE_OUT:-dist}" since commit 5708174 on 21 September — the variable is expanded by the shell before the flag is built, so the flag carried it rather than overriding it. Checked across four commits: 5708174, 88c90bc, 5d5e90e and 9507493 all pass the variable. And verified both ways just now: with ARCHIVE_OUT=dist-probe the gate names dist-probe and refuses; unset, it reads dist and passes 811 pages. So no gate here was grading a directory it had not built, and the sentence “every check run here today graded whatever was in that directory” is not true of this archive. THE UNDERLYING FIX IS GOOD AND IS KEPT. kit/tools/outdir.py makes the environment beat the flag for all four tools, which is right for the seven archives that still hardcode --dist dist and is a no-op here, since flag and variable already agree. The kit also gained the empty-build refusal this archive wrote for itself yesterday — “only 0 page(s) under … the build is unfinished or in flight” — which is the better home for it. The correction is only to the description of this archive, and it is published rather than amended because the commit is pushed and the tree is shared.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Commit faa05bc: “The kit also gained the empty-build refusal this archive wrote for itself yesterday.””
now THE KIT HAD IT EIGHT DAYS EARLIER AND IT CAME FROM FALCO. The refusal — “only 0 page(s) under … the build is unfinished or in flight. Refusing to pass.” — entered kit/tools/checkliving.py at c8548ee on 14 September 2026, “Two living-people guards become one, with three policies”, and the same day at 0f1e3f4, “Falco's living-people guard, generalised to all eight”. This archive's own no-site guard in checkregisters.py went in on 22 September. So the kit did not gain it today, and this archive did not write it first. THE LANDING SESSION REFUSED THE CREDIT I OFFERED IT and said so plainly rather than accepting a thank-you for a line it had not written, which is how the error was found at all. THE TWO GUARDS REMAIN THE SAME IDEA REACHED TWICE, eight days apart, by different archives against different faults — Falco's from the living-people rule, this one from a run that reported 205 failures over a site with no pages in it. That convergence is worth more than either of us having handed it over, and it is not a claim of provenance. Checked in the kit's own history before this correction was written, which is what should have happened before the commit message was.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “Anna Jacoba Kolbe married Francis McCabe on 13 July 1871 — one of two marriages between the Commandant’s children and the Landdrost’s”
now SHE IS NOT A DAUGHTER. SHE IS THE WIDOW, UNDER HER FIRST HUSBAND'S SURNAME, and this archive briefly held one woman as two people. South Africa, Civil Marriage Records enters ANNA JACOBA KOLBE; the archive already held ANNA JACOBA VAN DER MERWE, widow of J. G. E. KOLBE, remarrying Francis McCabe “about April 1871” on a Geni profile. Same groom, same year, three months apart — and a widow is recorded under her married name, which this archive already knew, having noted that the Free State estate index files her under McCabe for the same reason. THE TELL WAS ARITHMETIC, NOT PALAEOGRAPHY: a man does not marry twice in three months, and the duplicate stood on the page for a day before anybody added the two dates together. WHAT SURVIVES AND IS BETTER THAN WHAT WAS LOST: the marriage now carries a documented date, 13 July 1871, in place of a family tree's about April. The duplicate person is withdrawn, the two timeline entries are merged, and the marriage register names the widow. WHAT WOULD PUT IT BEYOND DOUBT: a civil marriage entry records each party's condition, so the unread image says “widow” or it does not, and settles it in a word. The second McCabe marriage is untouched — Jan Hendrik Kolbe and Rose Maria McCabe, 4 June 1872, remains two young people and not a widow. 27 SEPTEMBER 2026 — THE IMAGE IS READ AND IT SAYS THE WORD. The register enters her “ANNA JACOBA KOLBE, GEB: VAN DER MERWE”, 46, WEDUWE, and him WEDUWNAAR. The withdrawal was right, and it is now held on the document instead of on the arithmetic that found it. WHICH IS WORTH SAYING PLAINLY: THE ARITHMETIC WAS ENOUGH TO ACT ON AND NOT ENOUGH TO REST ON, and four days passed between the two.
how A document read wrongly, or a conclusion drawn past what the document says.
was “A civil marriage entry gives ages, occupations, fathers' names and witnesses — so the image would name the fathers”
now IT DOES NOT, AND THIS ARCHIVE SHOULD NEVER HAVE SAID IT WOULD. The claim was written into the work list, into two people's records, into two marriage rows and into a timeline entry, and it was the whole reason for wanting the images. THE FREE STATE CIVIL MARRIAGE FORM OF 1869–72 HAS NO PARENTS' COLUMNS. Read at the page, its columns are: No., Datum, Volle naam en toenaam, Ouderdom, Staat, Ambt of beroep, Woning ten tijde des Huwelijks, Huwelijksgeboden of Licentie, Toestemming — then the parties' own signatures, the witnesses', and the minister's. CONSENT IS RECORDED BY RELATIONSHIP AND NOT BY NAME: the 1872 entry says “Moeder” and stops. WHERE THE ERROR CAME FROM is the shape of other countries' civil marriage registers, which do carry fathers — an assumption about a record set, made without opening one, and then repeated confidently enough that it survived being written down five times. WHAT IT COST AND WHAT IT DID NOT. No conclusion rested on it, so nothing has to be unpicked; but it stood as the justification for a piece of work, and a reason to do something is a claim like any other. WHAT THE IMAGES GAVE INSTEAD IS BETTER THAN WHAT WAS EXPECTED: a condition, an occupation, and a farm that puts three of these people in one household.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “Williem Johannes Kolbe of Panfontein was born on 9 July 1847 in the district of Philippolis”
now THE REGISTER SAYS 2 AUGUST 1847, AND FAURESMITH. The 9 July came from the Afrikaans Wikipedia article, which cites P. J. Nienaber, Afrikaanse Biografiese Woordeboek (1947), p. 139 — a page this archive has never seen, reached at two removes. The baptism entry (FamilySearch 1:1:Q2HL-BYCG, Free State Dutch Reformed registers, entry 18) gives born 2 August 1847, baptised Fauresmith 12 March 1848, of a father the register abbreviates to “John Kolbe” and of Anna Jacoba van der Merwe. A REGISTER WRITTEN SEVEN MONTHS AFTER THE BIRTH BEATS A DICTIONARY WRITTEN A CENTURY LATER, and the archive now carries 2 August. WHAT IS NOT CORRECTED IS THE PEDIGREE — Nienaber's parents are exactly the register's, which is the dictionary being vindicated on the thing that mattered and wrong on a date. THE DISCREPANCY IS LEFT VISIBLE rather than tidied: four weeks is too small to be a different man and too large to be a slip of the pen, and somebody may one day know why. A NOTE ON THE ATTRIBUTION BELOW THIS ENTRY. The register's abbreviation of the Commandant — John Kolbe — is also, letter for letter, the whole held name of a different man in this archive: the Settler's son John Kolbe, who has nothing to do with it. The attribution tool matched him, correctly by its own rule and wrongly in fact, and he is refused in corrections-who-deny.json. A quotation is not a mention, and a tool that cannot tell the difference needs a human to say so once.
how A document read wrongly, or a conclusion drawn past what the document says.
was “Two collaborative trees disagree about Frederik Fortunatus Kolbe of Bloemspruit's birth year, so neither settles anything”
now TRUE, AND THE REASONING UNDER IT WAS WRONG. This archive wrote on 20 September that the disagreement between Geni's 1857 and the FamilySearch tree's 1859 showed the two were not independent witnesses. The opposite turns out to be the case: each tree is following a different document. 1859 is his gravestone; 1857 is the age of 73 on his civil death registration. THE TREES WERE THE ONLY HONEST PARTIES IN THE ROOM — they disagreed because their sources disagree. AND GENI'S DATE IS A SPLICE: “22 October 1857” takes the day and month from the stone and the year from the death form, so it reproduces neither document while looking more precise than both. THE RULE THAT SURVIVES, SHARPENED: where trees differ, do not ask which tree is right — ask what each one is copying, because a compiled date can be a composite of two sources and match neither. AND THE FIRST DRAFT OF THIS ENTRY WAS FILED AGAINST THE WRONG MAN. Written with the bare name, it was attributed to Frederik Fortunatus Kolbe (1839–1929), the Settler's tenth child — a different man by twenty years, whose page would have carried a correction about somebody else's gravestone. The archive holds four men of this name and the tool prefers the longest, so the fix was to write the longer one. Caught the same morning, by reading what the tool printed rather than that it exited 0.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “Of three Johan Casimir Kolbes in this archive's own register, not one is joined to this house”
now ONE OF THEM WAS THE SETTLER'S SON, AND THE EVIDENCE WAS ALREADY PUBLISHED HERE. Written on 21 September while resolving the Casmier spelling, and correct in its reasoning: a shared surname is not a relationship, and the three men were index rows rather than witnesses. BUT THE SECOND ROW — JOHAN CASIMIR KOLBE, married ELIZABETH GEERTRUIDA PIENAAR — IS HIM. His banns were called at Colesberg on 1 May 1867, the town his father's death notice was signed in and his mother died in that year; their children include a George Augustus and a Margaret Dewning. WHAT WENT WRONG IS NOT THE RULE BUT THE TENSE. Not joined is a statement about what has been done; not joinable is a statement about the world. The sentence was written as the first and read ever afterwards as the second, and so the row was never worked. THE FIX IS A HABIT, NOT A GATE: when this archive refuses a lead for want of a link, it should say what would supply the link — here, the mother's name, which is exactly what found him six days later.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “A search set published twice, because a script saved one file and then crashed on the next”
now THIS ARCHIVE PUBLISHED THE SAME SEARCH SET TWICE AND NOTHING SAID SO. A script appended the FamilySearch signed-out set to searched.json, saved it, and then died several lines later on a max() over a work-list column another session had typed as a string. The crash message was about the work list; the damage was in searched.json, already written. The corrected re-run appended the set again, and the page carried it twice through two commits. WHY IT IS A CORRECTION AND NOT A TYPO: /searched/ is a record of what work was done, so a set printed twice says this archive searched the same ground twice on the same day. That is a claim about the work and it is false. THE GENERAL FAULT, which will happen again: a script that writes several files is not atomic, so a failure in the last one leaves the earlier ones committed — and re-running it is not idempotent unless somebody made it so. Caught by an assertion in the very next script, which expected two matches and found three; a gate now refuses two sets sharing a heading and a date.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “John Augustus Barry I and Margaretha Kolbe married about 1850 at Burghersdorp”
now THEIR ELDEST SON WAS BORN IN 1847, AND THIS ARCHIVE DREW HIM UNDER THAT MARRIAGE. The 1850 rested on one word — “probably” — written by their son R. P. Barry on his father's death notice in 1915, sixty-five years after the event. The same son had written “Unknown” on his mother's notice three months earlier, which this archive quoted elsewhere as a sign of how little he knew, and then used his guess as a date anyway. Meanwhile people.js said “married 1846” the whole time. Two of this archive's own files disagreed, and the marriage page published the wrong one. WHAT IT IS NOW: 23 FEBRUARY 1846 at COLESBERG, marked inf because the exact date and place come from Pama, reported in a letter of November 1999 that had been in David's files for twenty-seven years. THE BOOK IS NOT WHAT MAKES IT SAFE — this archive does not let a book overrule a page. 1850 is refuted by arithmetic on the archive's own data, and 1846 was already here. HOW IT WAS CAUGHT MATTERS MORE THAN THE DATE. Not by a gate, not by an audit, but by a stranger's letter arriving at a page nobody had re-read. There is a gate now — child-before-wedding refuses a household whose marriage postdates a child drawn under it, and it fired on this row and on James Henry Barry, born 1848, before the fix.
how A conclusion that does not follow — the documents read correctly and the inference drawn wrongly, or a control that tested the wrong thing.
was “William Frederick Thomas Barry was born on 29 September 1861”
now THAT IS HIS BAPTISM DATE, AND THE REGISTER RECORDS NO BIRTH DATE FOR HIM AT ALL. Entry 79 of the Graaff-Reinet baptism register is headed “Gedoopt 29 Sept. 1861” — baptised. The entries on either side carry a birth date and his does not: 78 is geb. 24 April 1861, 80 is geb. 16 Aug. 1861, and 79 is a name with no geb. This archive took the one date on the entry and filed it as a birth, which is the commonest error there is in a baptism register and the easiest to make when an index labels its single date Event Date. What is actually known is that he was baptised on 29 September 1861 and born at some point before it. AND A SECOND THING GOES WITH IT: this archive reasoned from “born 29 September 1861 at Graaff-Reinet” that the household had returned to Graaff-Reinet between Burgersdorp and Dordrecht. The godparents are the mother's sister and her husband. A child is christened where its godparents are; a baptism place is where the font was, not where the family slept.
how A document read wrongly, or a conclusion drawn past what the document says.
was “The people register follows twelve lines, and one of them is called Testimony”
now IT IS A CONFIDENCE LABEL SITTING IN THE LINE FIELD. One person — a living Booyzen, held on testimony — carried l: "Testimony", and Testimony is this archive's own word for s: "lore", three fields away in the same record. So the reader was offered “Testimony (1)” in the line filter on /people/, between Daschner (15) and Lerena (1), as though this archive followed a family of that name. NOBODY WAS MISDESCRIBED AND NOTHING WAS CLAIMED ABOUT HER — the line field only groups — but the register's own list of the families it follows was wrong, and that list is read as a statement. IT WAS FOUND SIDEWAYS, while counting the l values to answer a different question entirely: what the Record Atlas should be told this archive's families are. A field whose values are never counted can hold anything. The line is now Booyzen, which is her surname and was always the answer.
how A mistake in the archive's own working — a bug, a bad build, or a word used loosely.
was “The Frederik Fortunatus Hofmeyr Kolbe in the DSAB's printed index is probably a misreading”
now IT IS A BIOGRAPHICAL ENTRY, AND IT NAMES HIS PARENTS. Row 173 met the name in a user tree, found the same string in the Dictionary of South African Biography's index, and wrote that it was probably a misreading — reasonable caution about an index line nobody had opened. The entry is at Volume V, page 419: born and died at Wakkerstroom, 27 September 1919 and 18 June 1969, an agricultural extension officer, the son of Frederick Fortunatus Kolbe and his wife Anna Margaretha Catharina Moolman. The dates match the tree to the day and the dictionary adds the two names the tree did not have. THE CAUTION WAS RIGHT AND THE GUESS INSIDE IT WAS WRONG, and the difference is that the index was treated as a thing to explain away rather than a thing to open. Still nothing is joined: a published entry about a man at this archive's Kolbe place is a lead, not a descent.
how A document read wrongly, or a conclusion drawn past what the document says.
Nothing at that level.
What the pattern says
Three of the nineteen are the same mistake, and it is the one worth naming. The archive searched what it happened to have, found nothing, and reported the absence as a fact about the world. "Not one miner among them" came from six occupations. "The one time the form worked" came from five death notices. "They are not this family" came from a geography the archive had never actually tested.
A list is not a population. Every negative on this site now carries the scope it was run over, and where the scope is small the page says so. What was searched →
Three more were the reverse: a search run over part of the ground and written up as the whole. One province out of nine. Twenty-two documents out of forty-two. A register that turned out to cover a different colony entirely.
And one, the most recent, is the worst kind: a fact this archive manufactured from a reference number and then quoted back to itself. An estate file numbered O/2899 was dated 1885; the archive wrote 1885 into a man's life as a death date, and then cited the death date as though it were evidence for the attribution. How that unwound →
Every correction is also in the log of what changed, by the day it happened, with what was being worked on at the time.