209 items, 68 of them open
The work list
What is being done here, as opposed to what is known.
Why this page exists
Every other register on this site answers what this archive knows. /searched/ holds every query ever run, /open-questions/ holds what is not known, /errands/ holds what can only be got through a door, and /corrections/ holds what was got wrong. None of them answers what is being DONE — and so for weeks that lived in a chat window, which is the one place that does not survive the session. This page is that list. It is deliberately not a list of findings. It is who is doing what, who is waiting on whom, and what was decided against and why.
The other registers on this site are about the family. This one is about the work, which means it is the only page here that is mostly about this archive's own limits — what it is waiting on, what it cannot reach, and what it set out to do and had to strike because the plan rested on something untrue.
An item here with no owner would be wallpaper, so the build refuses one, and an item open for six weeks is marked as such rather than left to look like something raised this morning. FROM ITEM 24 THIS BECOMES A ROADMAP RATHER THAN A DIARY. Those items were not raised by anything going wrong; they are the work this archive can see from where it now stands, in four groups and in the order it intends to take them. TEN DEPOT CAMPAIGNS THAT HAVE NEVER BEEN RUN — whole indexes this archive has taken single files out of without ever sweeping. SEVEN CHECKS, all of them owned by a machine, because every convention that has held here held when a build refused and every one that drifted was written down and watched by nobody. SIX PIECES OF PUBLISHING DEBT, which is what it is called when a page argues something the evidence has since outgrown. AND THREE THAT LEAVE THE BUILDING. A SECOND ROADMAP BEGINS AT ITEM 51, raised the same evening, and it is mostly the cost of a good day: nine research campaigns opened by things found on 15 September, six more machine checks, seven pages that now argue less than the evidence supports, and four things for the estate and the post. A DAY THAT FINDS A GREAT DEAL ALSO MAKES A GREAT DEAL OF WORK, and pretending otherwise is how a list becomes a monument. A list this long is a risk in itself: the failure mode is not forgetting an item but stopping reading at item 12. Anything here that is still untouched in six weeks should be struck rather than carried.
The list
1. Done — finished and published this archive
Sweep the Transvaal education series (TAB, source OD) for every surname and farm this archive holds
One accidental query — BOOYZEN AND MARICO — returned a farm school at Kwaggafontein with a JGL Booyzen on eight of its forty-eight files, and a second school on this family's own farm at Vaalbank. Neither was looked for. The series has never been searched for BARRY, MOUNTJOY, DASCHNER or KOLBE, nor for Brakfontein, Koedoesfontein, Driefontein, Zwartruggens or Goedehoop. What was found 15 September, paused: NAAIRS answers «the connection to the Search Manager Server is inactive». Not a negative about anything — an outage. Done, 15 September. KOLBE returns one — G.A. Kolbe signing school declarations at Martinus Wesselstroom, the old name for Wakkerstroom, in 1899. BARRY returns nothing. And the farm half was abandoned as void: Brakfontein 77, Koedoesfontein 188, Goedehoop 41, all other people's schools. A farm-name sweep without a surname is not a search.
2. Done — finished and published this archive
Finish reading the two school files — 28 of 72 Kwaggafontein documents, 30 of 54 at Vaalbank
The unread half of Vaalbank is where the Ds. Booysen question lives: whether that abbreviation is a pair of initials, David Schalk, or the Dutch for a minister. The index will not settle it; the file might. Parked by the same outage. Done, 15 September. All fifty-four Vaalbank documents read. The 1899–1900 files are J.P. Louw alone and no further Booysen appears — so DS Booysen is on exactly three of the fifty-four. And the school simply continues: Vaalbank No. 116, Lichtenburg School Board, established 1905, establishments to 1937, a cleaning allowance in 1929 and 1937, transport schemes to 1954. The farm school on this family's own ground ran for at least forty-four years. Kwaggafontein was already complete at 48 of 48 once the district filter is applied — the 72 included other people's Kwaggafonteins.
3. Done — finished and published this archive
Sweep the eGGSA newspaper transcriptions for the four surnames that have never been put through them
KOLBE was swept properly for the first time today: 43 hits, read one at a time, and it produced Francis McCabe, the location of the farm, a son of the twelve, and the Settler's mission school. BOOYZEN, BARRY, MOUNTJOY and DASCHNER have never had the same treatment. The search endpoint is now scriptable, so this is an hour's work rather than a day's. Today's run Done, 15 September. Daschner returns nothing in the whole collection; Booyzen five and Booijsen one, almost all of it other families; and MOUNTJOY gave the thing this archive did not expect — a man who lodged in James Mountjoy's house in January 1822 and recorded that he was discharged from the 21st Light Dragoons. What it broke
4. Done — finished and published this archive
Find Francis McCabe's own estate at the Free State depot
He was Landdrost of Bloemfontein and died on 23 February 1872 leaving «a widow, four sons and two daughters, besides a large family of step-children». Those step-children are the Commandant's twelve. His file would name them — reaching the same list from the opposite side to VAB MHG M777, which is not readable online. Parked by the same outage. Done, 15 September, the moment NAAIRS came back. VAB MHG C129 exists and names both his wives — a first marriage to Maria Petronella Elisabetha Antoinetta Raath that nobody here knew about. His insolvency file says outright that it holds no death notice. And his widow's own entry names both her husbands, so the remarriage is in the archives and not only in a newspaper. What it gave
5. Done — finished and published this archive
Bring /open-questions/ up to date — §0 is a day behind and the letters section is three days behind
§0 still ends «the next step is the Graaff-Reinet baptism register for 21 September 1788». It does not know that the 1981 revision was confirmed this morning, that no Gerhardus Lodewicus married a Magdalena de Beer, or that the disputed name sits nine lines lower on the same sheet as a son. The page also still calls NAAIRS «the single most important unopened door», which stopped being true three weeks ago. Question 0, as it now stands Done, 15 September. §0 now carries the 1981 confirmation, van Zyl's failed marriage search, the son on line 12 and the closed FamilySearch route. §12 is rewritten around the Light Dragoons and moved from Blocking to High. The letters section says Pretoria answered.
6. Done — finished and published this archive
Read the rest of de Villiers's volumes 2 and 3 for the surnames not yet looked up
Volume 2 gave Anna Maria van der Merwe her baptism and six siblings; volume 3 gave Adriana Christina Viljoen her parents and made her David Schalk's first cousin. Not yet looked up: the van der Merwe siblings' own descendants, the de Beer entry, the five Dormehl mentions, the Oliviers, and the six Slier entries in volume 3 which may be Anna Catharina Slier. What volumes 2 and 3 gave Done, 15 September. The six Slier entries turned out to be a whole family: Willem Slier of Hamburg and Anna Catharina Fortman, married 1780, and their daughter baptised 14 October 1787. Two pages on, the Pietersens — which showed this archive had been running a husband and his son under one name. Dormehl is in volume 2 and is ninety years too early to connect. What it gave
7. Done — finished and published David
Engage the private researcher at Pretoria — done: Anne Brest, invoiced, going Tuesday 22 September
The National Archives answered that all six MHG files are bound and cannot be scanned and emailed, and put a private researcher who specialises in estate files on copy. She was written to today and asked what she charges, how, and whether the work is worth doing at all, with an offer to start with one file. This is money, so it is David's to decide and not this archive's. If she charges per page, MHG 3734/46 is the one to buy first: Catherine Mary Sophia Daschner's notice should name the parents this archive still cannot prove. The files DONE. Anne Brest (`digitalphoto@icon.co.za`, annebrest.co.za), a photographer-genealogist in Johannesburg, referred by Ms Mavis Xaba of the National Archives Reading Room because the files are bound and cannot be scanned and emailed. She quoted R2,850 for the first file including the drive to Pretoria, then by the work done and capped — and she volunteered the cap herself, unasked. Invoice No 463 of 16 September: seven files, R11,400 plus R200 for the South African transaction deduction, the seventh (WEAR, MHG 10245/69) at no charge because it is one page. She goes on Tuesday 22 September 2026. Her email of 19 September asks for confirmation of payment or a POP; that is David’s and nobody else’s.
8. Done — finished and published this archive
Sign in to FamilySearch
Four pieces of work are parked behind one sign-in. The Burgersdorp baptism register sweep is stopped at image 604 of 818; the whole-film nameless sweep has not begun; three Kolbe children and Angeline of Wakkerstroom are unfollowed; and the phthisis reading on Henry James Booyzen's death record is unchecked. Nothing else on this list is waiting on so little. 16 SEPTEMBER: SIGNED BACK IN, AND THE FIRST OF THE FOUR IS ALREADY DONE. The two missing Barry children of open question 13 fell in under an hour — Margrith Louisa Maria, baptised Burgersdorp 5 April 1852, and William Frederick Thomas, born 29 September 1861 — not to a name search but to the relationship search. Remaining behind this sign-in: the Burgersdorp hand sweep from image 605 of 818, the three unfollowed Kolbe children and Angeline of Wakkerstroom, and the phthisis reading. What it gave ··· DONE — David signed in on 16 September, and the account was used the same day for the Savoy catalogue search and the film check. ··· 17 SEPTEMBER — THIS ROW WAS CLOSED TOO EARLY AND THAT WAS THIS ARCHIVE’S SLIP. The sign-in is done. The four jobs parked behind it are not, and marking the row done buried them. They are now rows of their own.
9. Blocked — waiting on a person or an institution David
Send the Pietermaritzburg letter, or telephone first — the email address is in doubt
Six Natal estate files, drafted and deliberately unsent. The National Archives' own record gives pmbarchives@dact.kzntl.gov.za on an entry last touched in 2015, and kzntl.gov.za is a retired domain; other listings give pmbarchives@kzndac.gov.za. The postal address and the telephone, 033 342 4712, agree across every source. This archive has already posted one enquiry into a mailbox it could not confirm and will not do it twice.
10. Blocked — waiting on a person or an institution Western Cape Archives
MOOC 8/18.11 entire — the 1781 inventory that names eleven enslaved people
The heaviest thing on this list. The index line names Silvia, Lena, Sara, Janna, Dorinda, Fortuijn, Meij and Februarij van de Caab, Dijk and Farla van Madagascar, and Achilles van Mallabaar, in the estate of Gerrit Booijsen at Drakenstein. TANAP transcribed these inventories in full and its database no longer resolves, with the Internet Archive down as well, so there is no screen route left at all. What is wanted is the document: every name and valuation, and the heirs — because if the heirs include a Jan, this archive's direct line runs through it. What the index line says
11. Blocked — waiting on a person or an institution Western Cape Archives
Two letters to the Western Cape Archives, sent and unanswered
The Mountjoy wills and the Colonial Office memorial went on 9 September; the Lerena death notice on the same evening. Six days, no reply. Not yet chased, because a reading room that has had one working week is not late.
12. Blocked — waiting on a person or an institution David
Decide whether a gendatabase.com subscription is worth buying
The boundary is now known exactly. Free: names, surname, date of birth, baptism date. Paid: baptism place, source, image, entry number, father, mother, parents. The free layer gives dates and the paid layer gives parentage — and this archive has most of its dates and is starved of parents. One specific question would be settled by one record: whether the Johanna Catharina Mountjoy born 28 January 1860 was baptised in the Barkly East district, which would make her the woman who died at Indwe in 1911. What it gave for nothing
13. Done — finished and published this archive
One address at the SANDF Documentation Centre bounced and has not been replaced
The enquiry about UDF service and pension records went to two addresses. sandfdoc@mweb.co.za returned 550 relay access denied; archive@dod.mil.za did not bounce and has not answered. Half-delivered is recorded as half-delivered rather than as sent. Closed, 15 September, with a negative. Looked for again and there is still only the one address and it still bounces — it is what the Centre publishes on its own directory entries and on every guide that lists it. What is left is the telephone, the Schweikert Building, and archive@dod.mil.za, which has not answered in six days. As far as this can be taken from here.
14. Done — finished and published this archive
The letters are tracked in three places and no single one shows what is outstanding
Outgoing correspondence is split between prose on /open-questions/, entries in the research log, and files in the requests folder. Six enquiries are live right now, and reconstructing that took a search of an inbox rather than a look at this site. This list is the first half of the fix; the letters themselves still need one register. Done, 15 September: /letters/, eight enquiries with what came back, linked from the menu.
15. Struck — founded on a mistake, kept rather than deleted this archive
Find the 21 September 1788 baptism through the FamilySearch Dutch Reformed collection
Founded on a mistake about the collection and kept here rather than deleted. The plan was to locate the register holding Petrus Jacobus Booijsen's baptism and read it, which would name his father in a contemporary hand and end the oldest question on this site. The Baptism Finder was asked which registers exist for Paarl (Drakenstein), Graaff-Reinet and Stellenbosch: 54, 48 and 55 — and sorted earliest first, the first row in every town is 1843. It is the post-1842 church archive series and it does not reach the eighteenth century at all. The question is unchanged and the route never existed. It is a Cape Town errand now. Why the route closed
16. Struck — founded on a mistake, kept rather than deleted this archive
Read the seven Marico war-compensation files at a screen
Two days were spent treating these as a lead that could be worked online. All seven detail records were finally pulled today and every one reads, in full, «CLAIMS FOR COMPENSATION BURGHERS. MARICO. XX BOOYZEN.» No remarks, no farm, no forename, no amount. The index line is the whole record and no screen anywhere will improve it. Struck as a screen task and moved to /errands/, which is where a thing belongs when the only remaining step is a door.
17. Blocked — waiting on a person or an institution David
Chase the London tailor into a directory through Google Books
Whether John Godloff Kolbe of the 1844 obituary was at Brunswick Square or at 34 Conduit Street is still open. The books API returns HTTP 429 with a daily quota of zero for anonymous callers, so that route is shut and a different one is needed — archive.org full text, HathiTrust, or londonancestor.com. One thing did come out of it and is not being claimed: a list of 234 London master tailors who met over the Combination Act on 22 December 1799 includes «KOLBE, JOHN, 18, BROAD-STREET, SOHO». Right trade, right city, right age, and Soho was the German quarter — but a John Kolbe in London is not yet this John Kolbe. 15 September, the other routes tried and closed. HathiTrust is behind Cloudflare bot protection and this archive will not go around that. archive.org holds no London trade directory earlier than 1920. The 1810s-to-1840s directories are on the paid genealogy services and at the Guildhall Library and London Metropolitan Archives, and nowhere free. Now genuinely a door question, and the door is in London — which is where WO 97 is too, so the two travel together. 15 September: it travels with the Kew errand now. Every free route is shut and the volumes are at the Guildhall Library and the London Metropolitan Archives. Anyone going to Kew for WO 12/1450 (this row said WO 12/1430, which is the 20th and 21st Dragoons for 1867–68) is a mile from both.
18. Done — finished and published the kit and estate
Tell the estate that the FamilySearch Dutch Reformed baptism films begin in 1843
This belongs to every archive here, not to this one. Any archive working Cape Dutch Reformed registers through FamilySearch has the same ceiling and may, like this one, be reading its silence as absence rather than as the edge of a filming project. It explains retrospectively why weeks of work on this family produced nothing older than the 1840s. The originals exist; they are at Cape Town. Done, 15 September — written out in `notes/for-the-estate.md`, with the register counts town by town and the reason it is the expensive one: it does not announce itself. A search returns nothing and looks like an absence; it is an edge.
19. Done — finished and published the kit and estate
Offer the estate two search techniques worked out here
Both are general and neither is written down anywhere shared. NAAIRS can be driven end to end from a page — the rotating token, the database selector, the POST that must carry its AND operators explicitly, and the detail endpoint — which turns a frameset nobody can script into a few seconds per query. And the eGGSA newspaper index is an Elasticsearch behind a widget: its query shape takes a fragment count, so one request returns every mention in every transcript rather than the one the widget draws. Done, 15 September — both written out in `notes/for-the-estate.md`, with the actual request shapes. The NAAIRS note carries the trap that cost this archive a day (two keywords are ORed unless the operator is sent) and the eGGSA note carries the opposite caveat (that engine has no AND at all, so paired negatives there are void).
20. Done — finished and published this archive
The archive-kit pin was four commits behind and this page is why we noticed
This archive was pinned to b16bea79 and the work list component, its checker and the estate roll-up were added after it. Diffed before bumping — seven files, 658 insertions, nothing removed and nothing this build depends on touched — then moved to 60aad580. Recorded because an archive silently behind on the shared kit is the same failure this page exists to prevent, one layer down.
21. Next — ready to start David
Reply to Pieter van Zyl and Riël Smit on WikiTree — they answered the oldest question on this site and have not been told what it produced
Riël Smit confirmed from the 1981 edition this archive does not own that Petrus Jacobus is still placed under Jan, and Pieter van Zyl searched for a marriage between a Gerhardus Lodewicus and a Magdalena de Beer, found none, and remembered from the notice itself that a child of Petrus Jacobus carried the name — which turned out to be nine lines lower on the same sheet. Two people did unpaid work for this archive today and neither has been thanked or shown the result. The reply is this archive's to draft and David's to post; posting is public writing under his name, not something this archive does for him. What they settled Drafted 15 September and sitting in the requests folder, unposted. It reports what their answers produced, says plainly what this archive has *not* done — the chart still shows Gerthardus Lodewikus at the top — and asks them the one question left: whether the pre-1843 Drakenstein doopregisters have been transcribed or filmed anywhere this archive has not thought of. It also gives something back, since both of them work these families.
22. Done — finished and published a machine
A build check that every event on the atlas carries a citation
The Marico entry carried «1898 — JGL Booyzen of Kwaggafontein, in the field-cornet's papers» for weeks. The papers were the education department's, and the line survived because nothing on the site pointed at it — no page cited it, so no page could contradict it. An event nobody can check is an event nobody can correct. The other registers here are already gated: the kit check, the living-person rule, and now the work list. The atlas is the one register with no gate, and it is the one that was wrong. The correction Done, 15 September. `tools/checkatlas.py`, in build.sh after links.py. It refuses a build where a place pinned approximate carries no «why», and warns where a place with events is discussed nowhere on the site. First run: eighteen failures, every one a farm pinned near its district with nothing saying so. All eighteen now state what the pin rests on and several say outright that it is weak.
23. Done — finished and published this archive
File today's leads into /open-questions/ and /errands/ so they stop living on this page and in a chat window
This list is for what is being done and it is already carrying things that are really research leads — which is how it would turn into a second, worse copy of the registers it sits beside. Nine of them came out of today and have no home yet: the Volksraad of May 1867; the twelve remaining Booijsen inventories at MOOC8; the eight Kolbe sons through the National Archives Discovery catalogue and the German Savoy RG4 registers; the Kolbe-Booysen household at Port Elizabeth; «little Jannie» of 1884, whom two papers give two fathers; Jephts of Leeuwberg, the man of colour the 1869 court named with property of his own; and the Settler's Bushman mission school at Koolskop. Each belongs in a register that already exists. Moving them is the work; keeping them here is the drift. Done, 15 September, and here is where each went. §15 of open questions is new and holds the Commandant's twelve children, with John Kolbe junior as the one document has reached and the thirty-one tree names explicitly refused. Errands gains the Volksraad minutes for May 1867 and Francis McCabe's own estate under the Free State depot, and the twelve remaining Booijsen inventories under Cape Town. Kew is a new depot on the same page. The Kolbe-Booysen household, «little Jannie» of 1884 and Jephts of Leeuwberg are written up in the documents and carried in what has been read, which is the register for a lead nobody is working yet.
24. Done — finished and published this archive
Sweep the Cape estate index (KAB MOOC 6/9) for all five surnames, not just the files already known
This archive has read individual MOOC files — Kolbe 1845, Montjoy 1840, Downing 1854 — by interpolating reference numbers across film. It has never swept the series. A death-notice index is the single richest thing at the Cape for this family's nineteenth century, and BARRY, MOUNTJOY, DASCHNER and SLIER have never been put to it at all. Errands Done for MOUNTJOY, 15 September. Forty-two in the Cape depot. It confirms the 1834 Colonial Office memorial from «the widow of J Mountjoy» is really there, adds three estates including the bride of a 1901 marriage found in the newspapers — and turns up a whole other Mountjoy family, labour agents and traders in the Transkei from the 1920s to the 1960s, with a limited company of their own. Not this family so far as anything shows.
25. Done — finished and published this archive
Sweep the Transvaal estate index (TAB MHG) for BARRY, MOUNTJOY and DASCHNER
Done thoroughly for Booyzen — 184 documents, enumerated properly after a filter bug — and for Kolbe. The other three surnames have had only the targeted lookups that produced the six files now with a researcher. Depends on NAAIRS answering. Done, 15 September. MOUNTJOY 39, including a farm actually called Mountjoy and a Mountjoy occupying land in Rustenburg in 1907. DASCHNER 12 — the whole surname — giving William Herman a wife, a child who died in 1969, and a Henry James Daschner.
26. Done — finished and published this archive
Open the Free State depot (VAB) properly — all five surnames
VAB was opened for the first time on 15 September and only for two references, K71 and M777. The Commandant's whole public life was in that province: bank director, Justice of the Peace, jury foreman, Commandant of the district. His widow died there in 1896 and her second husband was the town's Landdrost. A depot touched twice is not a depot searched. Done, 15 September. 146 Kolbe documents and thirteen McCabes. It produced the Settler's missionary career, a letter from his father in London, the resolution of the Kolbe–McCabe question, Downing as a forename across four generations, and six Kolbe siblings applying to a court together in 1931.
27. Done — finished and published this archive
Open the Natal depot (NAB) beyond the six MSCE references
The six Natal files in the unsent Pietermaritzburg letter were found by targeted lookup. The depot itself has never been swept, and two of this archive's Kolbes were born in the Transvaal and died in Natal — which is a migration pattern, not a coincidence, and it will have left more than six files. Done, 15 September, and it was worth it. Thirty-nine Kolbes, nine Mountjoys, four Booyzens, no Daschners. Three files have gone onto the Pietermaritzburg letter that were not on it when it was drafted, including the wife of a man already on the list and both estates of the Settler's tenth child and his du Toit wife. What it gave
28. Done — finished and published this archive
Sweep the eGGSA gravestone library province by province for all five surnames
Four provinces have produced stones — Middelburg, Wakkerstroom, Silverton, Piet Retief — and each was reached by following a single lead rather than by sweeping. The provincial subdomains each carry their own two-letter CGI, which this archive learned the hard way after publishing four false negatives from guessed script names. The technique is known now and the sweep has still not been run. Done, 15 September — through NAAIRS's GEN database rather than province by province, which is one search instead of nine and reaches the same collection. MOUNTJOY nought, DASCHNER nought, and the controls make those real: Barry 54, Kolbe 4, Booyzen 3. Three new Kolbe stones, and the three Booyzen ones turn out to be one household in one private cemetery.
29. Done — finished and published this archive
Indwe and Barkly East past 1910 — read, and mostly already here
The archive stops where its leads stopped. Willem Hermanus Booijzen was confirmed at Indwe in 1907, married there in 1908 as a MIJNWERKER at Dugmore Mijn, and a Mountjoy died in the same congregation in 1911. Nobody has read what the register says after that, and it is the one place a colliery family's whole decade is written down in order. 15 September: this is behind the FamilySearch sign-in, like the Burgersdorp sweep, and listing it as ready to start was wrong. 16 September: unblocked. ··· DONE, AND THE RESULT IS SMALL FOR A GOOD REASON. Thirty-six Barkly East entries dated after 1910 across five surnames, checked one by one, and nearly all were already held. What is new: the Indwe lidmate roll is the register that matters, not the baptisms; Johanna Catharina Mountjoy joined it on 23 July 1907, seven weeks after Willem Hermanus Booijzen was confirmed in the same congregation; and Catherine Mary Sophia Barry BAXTER, born 1883 at Barkly East. A Booysen cluster at Indwe 1925–1932 is recorded and claimed by nobody. The sweep
30. Blocked — waiting on a person or an institution David
Johannesburg street directories, 1905–1935
Named in open question 10 as the private-record route and never followed. Forest Hill, Mayfair, Turffontein and Booysens are all addresses this archive holds without house numbers, and a directory gives occupation and street together — which is how to test the testimony that the boys went to the mines. 15 September: not a screen question. Longland's directories exist for the 1890s and are sold commercially as reproductions; the 1905–1935 run is not freely online anywhere this archive can reach. It is the Johannesburg City Library or a bookseller, which makes it a purchase or a visit rather than an afternoon.
31. Blocked — waiting on a person or an institution Western Cape Archives
The Cape Rebel trial files for Barkly East and Wodehouse, June 1901
Open questions 1 and 2 both end here. The register gives R.P. Barry of Jerpoint, Fred J. Barry of Ivanhoe and D.W. Schoeman of Caerlaverock tried in one month in one division, with no birth dates and no parents — so identity is unproven and stays unproven until the files are read. They are at Cape Town. 15 September: it is a reading-room job and always was. The trial files are at Cape Town and are not indexed by name anywhere reachable. Listed as ready to start when nothing here can start it.
32. Blocked — waiting on a person or an institution The National Archives, Kew
The 1820 Settler papers themselves, CO 48, rather than a compiled transcript
Open question 9 names this and calls it Low priority, which was right when the only gain was a cross-check. It is worth more now: the settler roll entered George Augustus Kolbe as a CHEMIST aged 20 when he was sixteen, and this archive cannot tell from a transcript whether that age was sworn or simply written down — and a second age overstatement in the same year is either a pattern or a clerk. 15 September: CO 48 is at Kew, so it travels with the Dragoons errand and the London directories. Three London errands now, which is an argument for one trip rather than three letters.
33. Done — finished and published this archive
The Discovery catalogue at Kew for all five surnames, now that one errand is already there
WO 97 for the 21st Light Dragoons is going there anyway. The same catalogue covers Colonial Office correspondence, Admiralty, and the 1820 settler papers, and this archive has never searched a British catalogue for any of its people — it has only ever searched South African ones and then wondered why the English end was blank. Done, 15 September, and it paid for itself immediately. The catalogue indexes the regiment as «21st Dragoons (Light)» and not as this archive had been writing it, which is the sort of thing that turns a real record into a nil return. Four spellings of the surname searched across the WO series: two Montjoys in the name-catalogued discharge series, neither ours. A controlled negative, and it redirected the Kew letter before it was written.
34. Done — finished and published a machine
A check that every person marked `doc` carries at least one link to where the document is discussed
The atlas check found eighteen pins with no stated basis on its first run. The people file has never been checked the same way. A person marked *documented* whose record links nowhere is a claim with no visible evidence — the same failure as the Marico atlas line, in the register this site is mostly made of. Done, 15 September. `tools/checkregisters.py`, in build.sh. Every person marked `doc` must link somewhere. Proved by breaking it: removing the links from Gerthardus Lodewikus Booysen fails the build.
35. Done — finished and published a machine
A check that every corrections row's link resolves to a real anchor
Fifty-three corrections, each pointing at the page that explains it. Nothing verifies those anchors. A correction whose link has rotted is a withdrawal nobody can read, which is worse than not having published the correction at all. Done, 15 September. Same tool. Every corrections href must resolve to a real page and, where it carries one, a real anchor. Proved by pointing a row at `#no-such-anchor-at-all`.
36. Done — finished and published a machine
A check that every `blocked` row in what-has-been-read says what would unblock it
Eleven rows are marked blocked. A block with no named remedy is indistinguishable from an excuse, and this archive has spent the whole week learning that the difference between *searched* and *could not search* is the most important distinction it keeps. Done, 15 September. Same tool, as an advisory rather than a failure. It found two rows marked blocked with no way out named; both now say what would lift them, and one of them says plainly that it should not have been marked blocked at all and is left as it is until somebody re-reads the original search.
37. Done — finished and published a machine
A check that kin.js and people.js agree on who exists
On 15 September kin.js was right and people.js was wrong — kin had Hermanus Pietersen and his son as two households while people had them merged into one man with a long name. Nothing compared them. A check that every name in kin resolves in people, and flags a person in kin whose household implies a different shape, would have caught it the day it was written. Done, 15 September. Same tool. Every name in kin.js must resolve in people.js, disambiguating token included. Proved by misspelling Hermanus Pietersen. It would have caught this morning's real error on the day it was written. Writing it also caught a bug in itself: the first regex matched kin.js's `warn:` field and reported six sentences as missing people.
38. Done — finished and published a machine
An external-link check, because one source went off the air mid-session
`databases.tanap.net` stopped resolving between one search and the next, and it holds the full transcriptions of the inventory that names eleven enslaved people. This site cites external sources it does not monitor. links.py checks internal links only. A source that has gone dark should say so on the page that depends on it, rather than looking live. Done, 15 September, and deliberately not in the build: `tools/checkexternal.py`, run on demand, writing to `sources-state.json` so the site can say a source is dark instead of quietly linking into nothing. 34 hosts, four genuinely dark — including TANAP, and including `nationalarchives.gov.za` written with a dot in the middle, in the very row that ends «written down because the wrong address cost this archive a day». It only found them because it looks for hosts named in prose; the first version checked hyperlinks and found five hosts out of thirty-four. And its first answer was wrong in a way worth keeping: it called three live government domains dead, because a department can be alive with nothing but MX.
39. Done — finished and published a machine
A report of anything open for more than thirty days
The work list component already marks a stale row in the page. Nothing reports it — so a thing blocked in October will still be sitting there in December looking like a thing raised this morning, which is exactly the failure this list was built to prevent one layer up. Done, 15 September. Same tool. Anything open more than thirty days is reported with its age and the question «still true, or should it be struck?»
40. Done — finished and published a machine
A check that the archive-kit pin is not behind the kit's own head
This archive sat four commits behind and only noticed because a component it needed was missing. An archive silently behind on the shared kit is a convention drifting without anybody seeing it. A warning at build time costs nothing. Done, 15 September. Same tool, advisory, and it fired immediately: the pin had moved to d1e902ea while the kit's head was fd92b51c. Diffed before bumping — the only change outside `.DS_Store` was the kit's own work list, nothing under `kit/` — then bumped, which also picked up a component that filters on the state pills.
41. Blocked — waiting on a person or an institution Western Cape Archives
Give the people named in the 1781 inventory a page of their own
Silvia, Lena, Sara, Janna, Dorinda, Fortuijn, Meij and Februarij van de Caab; Dijk and Farla van Madagascar; Achilles van Mallabaar — and ten more in the 1831 inventory at Onder Roggeveld. They are currently a table inside a section about a search engine, which is the wrong frame and the wrong weight. Eleven names is the whole surviving record of eleven people, and this site gives a page to far less. Wait for MOOC 8/18.11 entire first, so the page is built on the document and not on a catalogue line. Still waiting, deliberately. It stays open until MOOC 8/18.11 is read, so the page rests on the inventory and not on a catalogue line. 15 September: marked blocked rather than next, which is what it always was — it waits on MOOC 8/18.11 being read, and that is a reading-room job.
42. Done — finished and published this archive
Rewrite /namesakes/ — three name-derivations are now documented and the page predates all of them
David Schalk comes in with Anna Maria van der Merwe: once in the whole first volume of the Cape genealogies, thirteen times in the van der Merwe section of the second. Willem Hermanus is Willem Slier of Hamburg plus Hermanus Pietersen, neither of them Booyzen blood. Judith Cornelia Theresia comes in from the Viljoens and is handed straight down. The page argues that names descend; it can now show three of them arriving. Done, 15 September. Two new sets — David Schalk, with the count across three printed volumes that makes it the only derivation on the page that can be put as a number, and Judith Cornelia Theresia, which identified a Johannesburg gravestone from a genealogical code. And the Willem Hermanus set has been corrected at its head: it used to open on a man who was two people.
43. Done — finished and published this archive
Rewrite the Slier paragraphs on /name/
That page argues that a spelling is whatever the clerk had. Slier, Schleher, Sher and Hier — five renderings of one woman across five sources — is the cleanest example on the site, and it only became explicable when the book gave her father as *van Hamburg*. The page does not know that yet. Done, 15 September. A section on /name/ setting out the five spellings of Anna Catharina Slier and the one word that explains them — *van Hamburg*. It is the same argument as the z, on a name where nothing is at stake and therefore nothing is being defended.
44. Done — finished and published this archive
Rework /women/ around what the week actually produced
Anna Catharina Slier married at nineteen, was widowed twice, and lived fifty-eight years a widow to ninety-eight. Anna Jacoba van der Merwe was widowed on her wedding anniversary with twelve children, remarried the town's Landdrost, and was widowed again fifteen months later. Both were bare names a week ago. The page was written when they were. Done, 15 September. A section on /women/ for the two of them, and Anna Maria van der Merwe's row in the table of wives now carries her baptism and her disputed parents. Both women were found sideways — one out of a surname lookup for an unrelated reason, one out of reading a man's obituary to the bottom.
45. Done — finished and published this archive
Put the Free State on the front page
The index still reads as a Cape and Transvaal archive. Since 13 September this family has acquired a Commandant, a farm two hours from Bloemfontein, a Landdrost, a jury, a bank directorship and a deathbed opinion about the Basotho boundary. The front page is the one register that has not moved. Done, 15 September. Bloemfontein is in the eyebrow, the Commandant has a card of his own, the atlas card says what the new gate does rather than quoting a place count that had moved, and «three enquiries are out» is now eight with one answered, pointing at the letters and the work list.
46. Done — finished and published this archive
A page for the sources that cannot be reached, and why
Scattered across the registers there are now half a dozen routes that are shut for different reasons, and the reasons matter more than the shutness: TANAP off the air, HathiTrust behind a bot check this archive will not go around, Google Books at quota zero, the FamilySearch church films beginning in 1843, MHG files bound and unscannable, NAAIRS intermittently down. Together they are an argument about what a distant researcher can and cannot do, and at the moment that argument is only visible to somebody who reads every page. Done, 15 September — as a section and not a page, at /sources/#shut, because a new page nothing links to is a page nobody reads and this belongs beside the host table. Six doors, six different reasons, and the argument they make together: almost none of this is anybody refusing anything. Two are undocumented technical edges, one is a physical property of old paper, one is a library defending itself from robots, one is a price, one is a server having an afternoon. The biggest obstacle to this archive is not access — it is that the edges of collections are unpublished, and an undocumented edge is indistinguishable at a distance from an empty archive.
47. Done — finished and published David
Write the Kew letter
The errand is specified — WO 97 and WO 12 for the 21st Light Dragoons — and nothing has been drafted. Kew has a copying service and a list of independent researchers, and the question it would answer has been marked Blocking on this site since it was written. The errand Drafted 15 September and unsent. And drafted *after* searching rather than instead of it: Discovery was worked first, which changed what the letter asks for. It no longer asks anybody to look in WO 97 — that has been searched by name for four spellings and he is not there, with a control showing the absence is real — it asks whether the reading is right and points at **WO 12/1450 (this row said WO 12/1430, which is the 20th and 21st Dragoons for 1867–68), the musters for the 20th and 21st Dragoons. 15 September: NOT SENDABLE BY EMAIL. Kew takes records and research enquiries through a web form and publishes no address for them. The domain has a mailbox; nobody is invited to write to it. Submitting their form is a different act from sending an email and is David's to authorise on its own terms, so the letter waits. ··· THE WRITING IS DONE — drafted 15 September, corrected 17 September when Discovery showed the piece number was wrong. Sending it is row 73**, and this row was sitting at `blocked` on work that had already been finished.
48. Next — ready to start this archive
Chase the Western Cape Archives if nothing has come by the end of September
Two letters went on 9 September. A fortnight is the point at which a polite second note stops being impatient, and this archive should send one rather than let the enquiry quietly lapse. The letters 21 SEPTEMBER — THE CHASE IS WRITTEN AND IS BEING HELD, at `requests/cape-archives-chase-september.md`. Not to be sent yet: this row's own condition is the end of September, and sending it on the twenty-first would be the impatience the row was written to avoid. Both requests are in one letter, because both went to the same address on the same evening. It offers to narrow the order rather than repeat it — the first page of each will, and the memorial's first page and signature — and it invites the reading room to close the enquiry by saying a reference is wrong, which for three references taken from NAAIRS is a real possibility and a complete answer.
49. Done — finished and published the kit and estate
Offer the estate a shared record of what is shut and why
Item 46 is this archive's own version. The estate needs one list, not eight — every archive here is hitting the same closed doors, and each is discovering them separately and at the same cost. The 1843 ceiling on the church films alone will have wasted weeks somewhere else. Done, 15 September — folded into `notes/for-the-estate.md` as a fourth section, because it is a request rather than a finding: which institutions cannot be emailed at all, and the caution that a government domain with no website is often perfectly alive.
50. Done — finished and published National Library of South Africa
MSB 751 — the letter arrived on the third day, and it is the first thing he wrote himself
Sent 15 September. Four photocopied pages holding a letter of 23 August 1830 from Johan Kolbe in London to his son at the Cape on his ordination — the only document this archive has ever found showing the two men in contact. Three things are wanted: the letter, whether it carries a London address at its head (which would settle Brunswick Square against Conduit Street after three free routes failed), and whether «LMS missionary» comes from the papers or from the donor. The last is the one that matters: the only other evidence for the mission work is a recollection written down forty-five years later, and whether the two are independent or one memory told twice is the difference between corroboration and an echo. What the catalogue says ··· ANSWERED IN TWENTY-THREE HOURS. Melanie Geustyn, 16 September: the manuscript collections are kept in off-site vaults, she has sent for MSB751, and it reaches her desk after 11:00 on 17 September. She has also sent for the BACKGROUND FILE on the collection — the library’s own paperwork on the 1985 donation, which this archive did not know existed and which is where “where is the original” and “who was Dr Jean Griffiths” would be answered if they are answered anywhere. Nothing to reply to: she has asked nothing and said she will come back. The letter and the answer ··· 18 SEPTEMBER — THE DOCUMENT ITSELF. Four pages, headed “London August 23. 1830”, signed “Your affectionate father, J G Kolbe”. No street address — so Brunswick Square is still open. But the LMS acceptance is in his own hand, independent of Montgomery’s memoir; Henry is dying of dropsy; Elizabeth is “Betsey” and lives in Edinburgh; Charlotte is named and stops being an inference; and “kind love from your SISTERS & BROTHER” dates the obituary’s “last remaining of eight sons” to 13 April 1832. The Background File holds nothing on the donation. The letter
51. Struck — founded on a mistake, kept rather than deleted this archive
Read John Montgomery's memoir entire — it is a serial and two instalments have broken two walls
«Forty-five years ago» put George Augustus Kolbe at a Bushman mission school at Koolskop; «How I arrived in Cradock in January 1822» put James Mountjoy in the 21st Dragoons. Those are two separate walls on two separate families, from one old man writing in The Friend in 1869, and this archive has read two instalments of a series it has never counted. Neither was found by searching for a Kolbe or a Mountjoy. What the two gave STRUCK, 15 September, on the argument rather than the work. Both finds came out of sweeping a surname across the whole newspaper collection, and those sweeps are done — Kolbe 43 documents, Mountjoy six, and Koolskop appears exactly once in the entire collection, in the passage already read. Reading a long memoir to find what a name search has already found is work that looks like diligence. The 1870 instalments carry a full list of the Salem party, which is valuable and is not this family's: Kolbe came out in Owen's. Searching for the ship instead produced a first-hand account of the crossing. The storm
52. Done — finished and published this archive
Write down the whole of Cradock in January 1822, because he named it
«At that time there were only about half a dozen inhabitants in the place» — and then he lists them: Solomon the constable, S.M. de Beer, the Rev. Mr Evans of the Dutch Reformed Church, Mr P. Gie, James Mountjoy, the two Messrs Ziervogel (one Secretary to Government), Mr Heilberg, Fieldcornet B.J. Vorster and Gert van Deventer. A complete census of a town, in a sentence, five years after this family married in it. This archive keeps saying it names the people documents leave out; here are nine of them and it has used one. Done, 15 September. Nine named inhabitants, the constable who demanded Montgomery's pass in the street, the Landdrost four miles out at Driefontein across the Fish River, and the warrant already issued. The only picture this archive has of the place its people were living in.
53. Struck — founded on a mistake, kept rather than deleted this archive
Sweep NAAIRS for DOWNING as a forename, across every depot
Six documents found by accident show it running four generations and three provinces, ending with a Pieter Dowing Kolbe Lubbe in 1959 — surname Lubbe, both Kolbe and Downing reduced to forenames, a letter worn off on the way. Nobody has actually searched for it. Margaret Downing signed her marriage allegation with a mark in 1819; her name may be the most durable thing about her. STRUCK, 15 September. A bare DOWNING sweep returns 552 documents and the first thirty hold no Kolbe, Booyzen, Barry or Mountjoy. Downing is an ordinary English surname. The pairing is what works and the pairing was already run — KOLBE AND DOWNING returns sixteen, and those sixteen are where the four generations came from. The item was raised on the assumption that nobody had searched; somebody had, in the only form that works.
54. Done — finished and published this archive
Find the parent behind the six Kolbes who applied to a Free State court together in 1931
Leonora Steenkamp, Margaret Downing Fick, Georgina Augusta de Wet, Jacobus Wynand Louw Kolbe, George Augustus Kolbe and Catharina Elizabeth Malan, all born Kolbe or Kolbe. Applications of that shape are usually about a parent's estate, so a VAB estate of about 1930 should have the father in it — and the index can be searched for one without reading the file. The application Done, 15 September, on an inference the file would settle. George Augustus Kolbe, d. 1905, and Hester Jacoba Jacomina née Louw, whose estate went under curatorship the same year — which is what happens when there are minor children. The mechanism that names them is a forename: one of the six is Jacobus Wynand Louw Kolbe, carrying his mother's maiden name. Their two estates are now on the errands page beside the application.
55. Done — finished and published this archive
Use the Transvaal and Cape estate finders on gendatabase, which have never been opened
Ten specialist finders were catalogued on 15 September and four were used. The two estate finders take a reference and a year and resolve it to the film — which is exactly the step that cost this archive a day when it was interpolating MOOC reference numbers across 2,223 images by hand. Done, 15 September — and they are not what the item assumed. The Cape and Transvaal estate finders take a reference and resolve it to a film; they do not search a name. Useful for turning a reference into an image, useless for finding a person.
56. Done — finished and published this archive
Pull the 1925–1931 voters' roll entries for Jacobus Nicolas and John Barry Booyzen
Both are on the list of six bound estate files now with a researcher in Pretoria. A voters' roll gives an address and a district — and one of the two is the man whose estate reference this archive has never been able to find on film at all. Cheap, and it might tell the researcher where to look. Done, 15 September, and the answer is that the address is paid for. Both men are on the roll for voting year 1925 and that year is the only free field. Employment, address, race, town, division, voting district, province and voting number are all behind the paywall. The address was the entire point of the item — it would tell the Pretoria researcher where to look. That is now a precise fact for the subscription decision rather than a general one.
57. Done — finished and published this archive
Open the Civil Deaths and Opposed Applications finders
Neither has been touched. Opposed Applications is the divorce index, which this archive has never searched for any of its five surnames — and a contested divorce file names a marriage, a date, children and often an address, for families that left no estate. Done, 15 September. Both are browse tools rather than indexes; their contents already reach this archive through the single search box. They did give two men nobody had: Henry John Booyzen, 1929–1966 — a different man from the Henry James born 1887, and the difference is one forename — and Maruis Booyzen, 1963–2011. Neither is placed.
58. Done — finished and published this archive
Open the University of Cape Town camps database at source
Three Booyzens are in it — Gert Jacobus, Gertruida Magdalena and Jan Gerhardus Lodewicus — and this archive has only ever seen them as index lines through two intermediaries. The database itself carries camp, farm, district, age and date of death, and the record pages are public. The caveat it publishes about itself matters too: it is built largely from death lists, so it records who died far better than who was interned and walked out. Done, 15 September, and it corrects an open question here. Of the three, only two died in camps. Jan Gerhardus Lodewicus, aged two, is recorded «died in camp: No» with no camp and no farm attached, and open question 3 lists all three as though they were the same kind of record. The other two gain farms, districts, a note that Gertruida's camp may have been Irene rather than Vereeniging, and a source — RS 25 Transvaal Death Lists, National Archives, Pretoria. Cause of death on all three is one word: *ziekte*.
59. Blocked — waiting on a person or an institution The National Archives, Kew
Ask who else of the 21st Dragoons stayed at the Cape
The regiment was in the Cape Colony from 1806 and left in September 1817 (this row said 1816 until 17 September). Men discharged locally and left behind are a cohort, and they are the men who married into the colony and whose origins nobody recorded — which is precisely the shape of this archive's oldest wall. If a list of them exists, James Mountjoy is on it and so are several other names this archive has met without explanation. 15 September: it is a Kew question and it travels with the Kew letter. Nothing reachable from here lists the men discharged at the Cape; the musters that would show it, **WO 12/1450 (this row said WO 12/1430, which is the 20th and 21st Dragoons for 1867–68), are not name-indexed and have to be read. Left blocked rather than pretended at. ··· 16 SEPTEMBER — PART OF THIS IS NOW ANSWERABLE AND PART IS NOT. FindMyPast’s WO 97 index for the regiment gives birth parish and county for 128 men who attested between 1793 and 1813 — England 95, Ireland 26, Scotland 2, Wales 2, one Swiss, one German. It does not carry a discharge place, so it cannot say which of them were discharged at the Cape. That still needs the muster books, WO 12**, or the discharge documents themselves.
60. Done — finished and published a machine
A check that every letter in the requests folder appears in the letters register, and the reverse
The register was built on 15 September from an inbox search because no single place on this site showed what was outstanding. It will drift back the first time a letter is drafted and not recorded, and nothing would catch that. Done, 15 September. In `checkregisters.py`: every `.md` in the requests folder must be findable in the letters register. Advisory rather than a failure, because a draft can legitimately exist for an hour before it is registered.
61. Done — finished and published a machine
A check that every searched.json set's link resolves, as corrections already does
Eighty-five sets, each pointing at the page that shows the working. The corrections check found this class of rot worth refusing a build over; the larger register has no equivalent and there is no reason for that except that nobody wrote it. Done, 15 September. Eighty-five sets, every `h` checked against the built pages and their anchors. A failure, not an advisory — the same standing as the corrections check it was modelled on.
62. Done — finished and published a machine
A check that people.js and people.json have not drifted
The header of the generator says they drifted once and four people vanished off the index. It is regenerated on every build, so the only way they diverge is a build that did not finish — which happened this afternoon, when an interrupted write corrupted a generated file and the next build failed on it. Done, 15 September. Counts and compares the names in `people.js` against `people.json` and names the first five in either direction. It would have caught the drift that lost four people off the index, and it would have caught this afternoon's corrupted generated file one step earlier.
63. Done — finished and published a machine
A check that a person whose own prose cites a tree is not marked `doc`
Several records say in terms «FROM WIKITREE, NOT FROM A DOCUMENT THIS ARCHIVE HAS READ» and this archive is careful about it by hand. By hand is how the atlas was careful, and the atlas check failed eighteen times on its first run. Done, 15 September, and it immediately raised five. A person marked `doc` whose own prose contains «not from a document this archive has read». All five are correctly marked — the record is documented and one *detail* inside it comes from a tree. So the advisory stands and the schema does not: there is no way here to mark a detail as tree-sourced rather than a whole record. Tuning the check to stop reporting them would be «a check with an exception is a check that will be excepted». [Raised as item 77]
64. Done — finished and published a machine
Make the external check name which pages cite each dark host
It reports that databases.tanap.net is gone and how many times it is cited. It does not say where, so acting on it means grepping. A check that reports a problem without locating it is half a check. Done, 15 September. Each host now carries the files that cite it. It paid at once: it showed that all four dark hosts are cited only from the registers that exist to record failure, and from the passages that say those hosts are gone.
65. Done — finished and published a machine
Run the external check again in a fortnight and keep both answers
One run is a snapshot. Two runs are a trend, and the question this archive actually wants answered is not «is TANAP down today» but «how fast is the ground moving under a family history that cites thirty-four hosts». Done, 15 September — the mechanism, not the second run. Each run keeps the previous answer, prints any host whose state has changed since last time, and keeps up to twelve dated snapshots. The second run is a fortnight away and the question it answers is not «is TANAP down today» but how fast the ground moves under a family history that cites thirty-six hosts.
66. Done — finished and published this archive
Rewrite /kolbe/ — the Settler was a missionary and his son ran a province
That page has him as an 1820 Settler with a medical practice who died at Colesberg. It does not know he was an LMS missionary, that his father wrote to him from London in 1830, or that his eldest son was Commandant of the Bloemfontein district, a Justice of the Peace, a bank director, a jury foreman and a member of the Volksraad who died saying the State had taken too much land from the Basutos. Done, 15 September. A new section on /kolbe/. The page had already *deduced* the missionary circuit from where the children were born; MSB751 says it in four words, and Montgomery's memoir says it from a third direction. Three things line up and none of them knew about the others. It also reconciles the occupations — chemist, pharmacist, medical practice — and adds that he was ordained in or about August 1830, which nothing here had said.
67. Done — finished and published this archive
Put today's two conflicts on /disputed/
Anna Maria van der Merwe's parents — the tree says David Schalk van der Merwe and Anna Elizabeth Louw, the printed genealogy says Willem van der Merwe and Hester Maria van der Merwe, and both cannot be right. And the Pietersen split, where this archive's own reading is the thing that was withdrawn. Neither is on the page for things this archive cannot settle. Done, 15 September. Two rows on /disputed/: Anna Maria van der Merwe's parents, where the tree and the printed genealogy disagree and the conclusion survives either reading, and «did Gerthardus Lodewikus exist?», which is the oldest question on the site and now has four things pointing one way and a chart that has deliberately not moved.
68. Done — finished and published this archive
Write «the controlled negative» into /method/
It has become the thing this archive actually does, and it is not stated anywhere as a method. A nil return is worthless until you show the search works — 95 WO 97 records for «21st Dragoons (Light)» before reporting that the man is not in it; the row filter that dropped thirty-six documents; the blank Name column; NAAIRS defaulting to OR; MOOC8 stopping before the Kolbes arrived; the church films beginning in 1843. Six instances in a week, one method, unwritten. Done, 15 September, at /method/#controlled-negative. Six nil returns from one week, each with what it actually meant, and the rule they produced: run a control and publish it beside the negative. When this archive reported that James Mountjoy is not in the discharge papers at Kew, it published ninety-five records for his regiment in the same series — because the number that makes an absence meaningful is not his, it is theirs.
69. Done — finished and published this archive
Put the Free State on the timeline
Ninety timeline events and the whole Bloemfontein decade is missing: a marriage of 1843, a lost promissory note on the Market Square in 1864, a jury verdict in 1866, a death on a wedding anniversary in 1868, a horse theft at Springfield, a farm advertised in 1870, a Landdrost dead on the Modder River road in 1872, a widow dying under his name in 1896. Done, 15 September. Twelve events, ninety to a hundred and two. The Bloemfontein decade entire — a marriage in 1843, a lost promissory note on the Market Square, a jury in 1866, a death on a wedding anniversary, a horse theft at Springfield, a farm advertised in 1870, a Landdrost dead on the Modder River road, a widow dying in 1896 under his name. And the build caught one of them: a place called «London» that this archive does not hold under that name.
70. Done — finished and published this archive
Check that /direct-line/ says what the evidence now says at the top
Jan Booijsen was entered as inferred on 15 September and the chart still shows Gerthardus Lodewikus, deliberately. That decision is defensible and it has to be visible on the page that makes the claim, not only in a work list and a document section. Done, 15 September. /direct-line/ now says at generation seven what the evidence says: that the 1876 clerk is the only thing holding the row up, that two editions of the printed genealogy put a different man there, that no marriage of a Gerhardus Lodewicus to a Magdalena de Beer can be produced, and that the row has deliberately not been changed because one source agreeing with its own earlier edition does not outrank a primary document signed by a son.
71. Done — finished and published this archive
Add MSB751 to /sources/, and the distinction it exposes
The Kolbe collection is in a library, not an archive — 5 Queen Victoria Street, not 72 Roeland Street, separate ticket, separate request. This archive has spent weeks treating «the Cape Town holdings» as one place. A manuscript in a library is invisible to every index it has been using, and it found this one by accident while searching for something else. Done, 15 September, at /sources/#library — and it turned out to be a bigger point than the item said. A library is not an archive, and for three weeks this project treated «the Cape Town holdings» as one place. The records are where the state kept them; the papers are wherever a descendant happened to give them. MSB751 was found by accident, and the section ends on the question that follows: how many other manuscript collections mention these families in libraries nobody here has thought to search.
72. Done — finished and published this archive
Write down what the 1869 memoir is worth, as a source type
Two walls in one week came out of one man's reminiscences in a newspaper, and this archive has no settled position on what that kind of source is worth. Montgomery is recalling forty-five years back; one of his claims has since been corroborated by a library catalogue and the other has not been tested at all. The gap between «a memoir» and «a record» is the interesting part and nothing here says where it sits. Done, 15 September. A memoir is a witness, not a record — no clerk, no form, no oath, no date of composition near the event — and what it has instead is circumstantial specificity. The test: does it claim something it had no reason to invent, and does anything independent touch it? Koolskop passes and the Dragoons does not, yet, and the two are written up differently on this site for that reason and not because one is a better story.
73. Blocked — waiting on a person or an institution David
Submit the Kew enquiry form, or hand the letter over
Written and unsendable by email: The National Archives take records and research enquiries through a web form and publish no address for them. Submitting a form on somebody's behalf is a different act from sending an email and is David's to authorise on its own terms. Their Live Chat runs Tuesday to Saturday and would answer the actual question — *would a man discharged at the Cape without a pension be in WO 97 at all* — in ninety seconds rather than three weeks. The letter
74. Blocked — waiting on a person or an institution David
Ask the estate whether any archive has a reader at Kew
This archive has one errand in London and is about to have two — WO 12/1450 (this row said WO 12/1430, which is the 20th and 21st Dragoons for 1867–68), and the trade directories that no free route reaches. Seven archives share this kit and at least one of them is working a British line. Asking costs nothing and the alternative is paying a researcher for a morning's work somebody may already be doing. 15 September: the asking is written into `notes/for-the-estate.md`, but this archive has no way to reach the other sessions — the shared work list is read by a roll-up, not a conversation. Passing it on is David's.
75. Done — finished and published the kit and estate
Tell the estate which institutions cannot be emailed at all
Kew takes a form and publishes no address. The South African Department of Sport, Arts and Culture has no website and only takes mail. The SANDF publishes an address that bounces. Every archive in this estate will hit these in turn and each will spend the same afternoon finding out, which is the argument for the shared record of shut doors already on this list. Done, 15 September — same file, same section.
76. Struck — founded on a mistake, kept rather than deleted this archive
Chase the National Library at a fortnight if nothing comes
Sent 15 September. Special Collections warn that some items are stored off site and ask for appointments, so a slow answer is expected rather than a bad sign — but an enquiry nobody chases is an enquiry nobody sent, and this archive has three letters out that it has never followed up. The letters ··· STRUCK, 16 SEPTEMBER — NOT NEEDED. She answered the next morning. The chase was written for an institution that had gone quiet, and this one did not.
77. Done — finished and published this archive
The schema cannot mark a single detail as tree-sourced, only a whole person
Raised by a check on the day it was written. Five people are marked `doc` and say in their own prose that some specific claim is «not from a document this archive has read» — Anna Maria van der Merwe is documented by a baptism of 1807 and her parents come from a tree; Adriana Christina Viljoen is named on a baptism of 1835 and her marriage date comes from a tree. Both halves are true and the status field can only hold one of them. The options are a per-claim marker, or accepting that a record's status describes its weakest load-bearing part. Until then the check reports all five every build, which is the right behaviour: tuning it quiet would be a check with an exception, and a check with an exception is a check that will be excepted. Done, 15 September, by adding the field rather than by tuning the check. A person now carries `tree:` naming which single claim is untested — «her parents», «the 1828 marriage date and place» — and the person page prints it beside the confidence chip, so a reader sees it and not only a checker. The check now fails a `doc` record that says a claim is undocumented and does not name it. Reporting the phrase and stopping there was half a check: it said something was wrong without saying what, five times, every build.
78. Running — a machine or a person is at it now this archive
FindMyPast — a full subscription, and it runs out on 16 October
THE CLOCK IS THE POINT. A one-month full subscription opened 16 September. Every British wall on this site has been beaten by a paywall or a bot check — Google Books at quota zero, HathiTrust behind Cloudflare, archive.org holding no London directory earlier than 1920 — and FindMyPast holds the record sets that answer nearly all of them. The campaign below is ordered so that if only half of it gets done, it is the right half. Anything still unharvested on 16 October is lost until somebody pays again.
79. Done — finished and published this archive
FMP · British Army service records — James Mountjoy, 21st Dragoons (Light)
The oldest blocking question on this site. His Cape death notice answers «names of the parents of the deceased» with NOT KNOWN and his 1817 marriage register rules a line through the column that should say where he came from. A memoirist who lodged in his house says he was discharged from the 21st Light Dragoons. Kew's own catalogue has no Montjoy of that regiment in WO 97, WO 119, WO 120, WO 121 or WO 25 — a controlled negative, since the same catalogue holds 95 WO 97 men of that regiment. FMP carries these series imaged, and its indexing is not Kew's. The question ··· 16 SEPTEMBER — RUN, AND IT IS A CONTROLLED NEGATIVE. FindMyPast indexes this regiment under four different strings holding different War Office series, and the obvious one — `21st Dragoons` — holds no WO 97 at all. All four swept in full, 1,110 records, and no Mountjoy under any spelling. Controls: WO 23/136 (1806–07) holds 18 men of the regiment and WO 23/147 (1817–26) holds 12. What it means and why it is not a surprise
80. Done — finished and published this archive
FMP · Britain, Directories & Almanacs — Johan Gottlieb Kolbe in London
Brunswick Square or 34 Conduit Street. An 1844 obituary puts the family at Brunswick-square; his own 1842 will and 1845 death record put him in Conduit Street, off Bond Street. Different parts of London and this archive cannot choose. Every free route failed. A trade directory of the 1820s–1840s decides it in one line, and it is the kind of thing FMP holds and nobody else digitises free. Also worth the same search: «Kolbe, John, 18 Broad-street, Soho» from a list of 234 master tailors who met over the Combination Act in December 1799. ··· 16 SEPTEMBER — ANSWERED, AND NOT BY A DIRECTORY. The Westminster rate books hold 121 entries under 24 spellings of his forename, 1800 to 1831: Broad Street North in 1800, which corroborates the 1799 master tailors’ list from an independent source, and Church Street South, St Anne’s Soho from 1811 to 1831. Brunswick Square is in none of them. The addresses
81. Done — finished and published this archive
FMP · London parish registers — the eight sons of John Godloff Kolbe
The 1844 obituary calls George Augustus «the last remaining of eight sons of John Godloff Kolbe Esq. of Brunswick-square». Seven brothers died before December 1844 and this archive has never found one of them. FreeBMD would not be scripted; the route was never tried anywhere else. George Augustus was baptised at St Anne's Soho, 2 February 1803, so the family was in that parish — seven baptisms and up to seven burials should be in the same registers. ··· 16 SEPTEMBER — FIVE OF THE EIGHT NAMED. Charles 1798, John Henry 1800, George Augustus 1803, Frederick Charles 1805, Charles Samuel 1807, with four sisters and mother Mary in every entry. Three sons are still missing and the gaps — 1808–12 and 1814–18 — are where they sit. The ten children
82. Done — finished and published this archive
FMP · Margaret Downing — born London, 23 February 1803
Read off a gravestone at Burgersdorp whose caption had stopped one line short for weeks. She has a birth date and a city and no parents at all. She signed her 1819 marriage allegation with a mark, swearing she was twenty-one at fifteen. A London baptism of that date would give her a father, and her surname is the one that runs four generations through the Kolbes as a forename. ··· 16 SEPTEMBER — NOTHING, CONTROLLED TWICE. 28 Downing baptisms in London 1801–1807 and 14 in Westminster’s own registers 1798–1810, and no Margaret in 1803 in either. The controls
83. Done — finished and published this archive
FMP · John Barry of Bilsdale, Yorkshire — break the conflation
Open question 6. One profile welds a Yorkshire farmer of Bilsdale, christened 1796 and buried at St Hilda's on 3 January 1853, to a John Barry who died at de Bruins Poort in the Cape on 8 January 1853, five days later. FMP has the 1841 and 1851 censuses for Holly Bower and the St Hilda's registers. Separate them, and see which — if either — is the father. ··· 16 SEPTEMBER — THE YORKSHIRE HALF DOES NOT EXIST. 5,934 Bilsdale burials are indexed, including the years either side of 1853, and not one is a Barry. 771 Barrys are buried elsewhere in the county. Open question 6
84. Done — finished and published this archive
FMP · Irish records — John Augustus Barry, «Irishman born in Ireland»
His 1903 death registration, made by a son who was in the room, answers question 5 «IRISHMAN BORN IN IRELAND» and gives his age as 84 years 2 months. His mother is recorded only as «born Neagle», with no Christian names and no address. The family speaks of County Cork. FMP's Irish holdings are the strongest of any service, and this is the one line of the family whose origin is claimed but never documented. ··· 16 SEPTEMBER — NOTHING IN THE CATHOLIC REGISTERS. Against 12,273 Cork baptisms for 1818 alone and 71 John Barry baptisms in Cork 1816–20, there are four John Barry baptisms in all Ireland to a mother named Nagle and none of them in his window. He may not have been Catholic. The controls
85. Done — finished and published this archive
FMP · the German Savoy Church and the RG4 non-conformist registers
Johan Gottlieb Kolbe was buried at the German Lutheran church of the Savoy on 26 November 1845, entry 55, grave 7, by the Revd Dr Steinkopff. The registers are TNA series RG 4/4625, 4626, 4628, 4631 for deaths and burials and RG 4/4627, 4629 for births and baptisms. This archive cited that burial from a tree and has never seen the register. If the family were Savoy Lutherans, the seven brothers may be there rather than at St Anne's. ··· 16 SEPTEMBER — THE REGISTER IS OPEN AND IT HOLDS SEVEN KOLBES. RG 4/4631, with one entry duplicated in RG 4/4630, a volume this note had missed. Three dead brothers dated, a daughter, and two wives, both called Mary. The baptism registers RG 4/4627 and 4629 are not on FindMyPast under any name tried — that is where the last three sons must be. The burials
86. Done — finished and published this archive
FMP · re-run the four South African surnames, now that the whole site is open
FindMyPast was used once, unsubscribed, and returned «12,984 for the surname and birth year alone» for Mountjoy — a number that means the search was never narrowed. It also produced one transcript reference this archive still cites without having seen the image, R_177176049779 for Petrus Jacobus Booysen's 1812 birth. Lowest priority of this campaign and it must still be run, because a transcript cited and unseen is exactly what this archive corrects other people for. ··· 16 SEPTEMBER — RUN, AND THE SET IS THIN. Barry 814, Booysen 126, Kolbe 37, Mountjoy 9, Booyzen 0, Daschner 0, almost all of it British records about South Africa. And the transcript reference R_177176049779 returns a 500 error on a live subscription — it does not open. Logged as a correction
87. Blocked — waiting on a person or an institution this archive
The Savoy baptism register — filmed, digitised, and locked to a library terminal
Five of the eight sons are named. Three are not, and the gaps in the Westminster run are 1808–1812 and 1814–1818. The German Lutheran burials are on FindMyPast; the baptisms, TNA RG 4/4627 and RG 4/4629, are not — a search for any surname at all with the keyword *German Lutheran* in that dataset returns nothing, so the set is absent rather than the name. Routes left: Ancestry, which images RG 4 in full; the Anglo-German Family History Society transcripts; or ordering the images from Kew. Also unresolved: the three women of the 1841 Conduit Street household — Marianne, Charlotte and Amelia A — whose census ages fall squarely in those same two gaps. ··· 16 SEPTEMBER — FOUND, AND SHUT. FamilySearch catalogue koha:176759, *Church records of German Lutheran Church, 1694–1840*, filmed from the originals at the Public Record Office. Film 008004264 — births and baptisms 1694–1840, marriages to 1754, burials to 1771, 537 images, and every volume has its own name index. And every one of the 537 images is padlocked, from image 1, on a signed-in account. Digitised and withheld: viewable only inside a FamilySearch Center or affiliate library. THE FIX IS A BUS FARE RATHER THAN A LETTER — and the index makes it a half-hour at the machine, not a page-by-page sweep. What it is
88. Done — finished and published this archive
The London Gazette and the Old Bailey — the loose ends from the Kolbe sweep
Found while sweeping the Kolbes and not opened: Henry Kolbe 1824 twice, John Gotlob 1824, John Henry 1824 twice, J. G. 1836, John Gbttlob 1836 and John Gotlieb 1845 in the London Gazette, plus John Henry Kolbe in the Old Bailey court records for 1819 (GBOR/OLDBAILEY/0399292 and 0399293). Gazette notices of that period are usually bankruptcies, dissolved partnerships or probate. John Henry is a brother — baptised 1800, dead 1832 aged 32 — so the 1819 and 1824 entries may be his, and this archive has not tested that. ··· 16 SEPTEMBER — THE OLD BAILEY HALF IS DONE. Trial t18190113-149 read in full, free, at oldbaileyonline.org: John Henry Kolbe, aged eighteen, deposing that “my father is a tailor, and lives in Church-street, Soho”. The trial The eight Gazette notices are still unread. ··· BOTH DONE, AND BOTH FREE. The Gazette holds five Kolbe notices, not the one this archive had: the Church Street partnership of father and son dissolved 12 May 1824, John Henry in the insolvent debtors’ lists that October and November, and a partner called Richard Davies dissolved in March 1845, eight months before the death. The five notices
89. Done — finished and published this archive
FMP · the 1841 and 1851 censuses for Holly Bower, Bilsdale — finish question 6
The burial half of open question 6 is now dead: 5,934 Bilsdale burials indexed, not one Barry. The census half has not been run properly — FindMyPast’s place filter is scored, not filtered, so a search for Barry in Bilsdale returns 112 rows sorted by relevance rather than the men who lived there. The right route is to page the census by registration district and enumeration district instead, or to search the household head and read the schedule. ··· DONE THE SAME DAY, AND IT PRODUCED THE MAN. The filter to use is `County` — “Yorkshire (North Riding)” cuts 3,255 Barrys to 52. Not one of the 102 Barrys in the two censuses is in the Helmsley district, and the only John Barry born 1796 in the whole Riding is a ship owner at Ship Yards, Ruswarp, Whitby, in the same house in 1841 and 1851. The two schedules
90. Done — finished and published this archive
FMP · the Boer prisoner-of-war list, WO 108/368
TNA WO 108/368, *List of Boer prisoners of war, numbers 1–32561, 1899–1902*, holds three men of that name taken in 1901: aged 53 of Pretoria and Boksburg, aged 41 of Vaalbank, Lichtenburg, and aged 33 of Kapsteinkop, Cradock. Two of those three places are this family’s own ground and one is the Vaalbank whose farm school this archive documented a week ago. It is a lead and not a finding — David Schalk is a van der Merwe name before it is a Booysen one. The whole Booysen and Booyens run in that series (153 men) has not been read. ··· RUN THE SAME DAY, AND THE KOLBES WERE THE FIND, NOT THE BOOYSENS. Six Kolbe men, each with a farm against his name — two of them called George Augustus, one at Wakkerstroom and one at Bethulie, and one Andries Frederick of LEEUWBERG, a farm this archive wrote down a week ago from a different document entirely. Five of the six are in the southern Free State. And a transcription rule worth more than any of it: this index writes Jan as “Ian”, so a search for Jan returns nothing. The list
91. Done — finished and published the kit and estate
Write the FindMyPast driver into the estate note before the subscription lapses
The harvest was not done by clicking. FindMyPast’s own GraphQL endpoint is `https://www.findmypast.co.uk/titan/marshal/graphql`, operation `searchResultsRecordsAndMetadata`, variables `{filters, order, pageNumber}`, twenty rows a page. Filter fields that work: `LastName` (with `variants`), `FirstName`, `EventYear` (with `offset`), `YearOfBirth`, `YearOfDeath`, `KeywordsPlace`, `Keywords`, `DatasetName`, `Regiment`, `SourceCategory`, `MotherLastName`. Two traps, both of which nearly produced a wrong answer today: place and year-of-birth filters are scored, not filtered, so they rank rather than restrict; and the same regiment is indexed under four different strings holding different record series. This belongs in [notes/for-the-estate.md] with the NAAIRS driver and the eGGSA query, because the next session will not have a subscription to rediscover it with. ··· WRITTEN, 16 September, into `notes/for-the-estate.md`, with the query document, the filter-field list, both traps and the transcript route.
92. Done — finished and published this archive
Sweep the London Gazette for the other four surnames, 1800–1900
The Gazette was searched once for Kolbe and turned up five notices where this archive had one — a dissolved partnership, two insolvencies and a business partner nobody knew about. It has never been searched for Mountjoy, Booysen, Booyzen, Barry or Daschner. It is free, it covers 1665 to now, and its Atom feed at `/all-notices/notice/data.feed?text=…&start-publish-date=…` returns the OCR text with the term highlighted. It carries insolvencies, bankruptcies, dissolved partnerships, unclaimed estates and colonial appointments — all of which this family generated. ··· RUN THE SAME HOUR, AND EMPTY. Montjoy 14 — all of them the chancery cause list *Lord Montjoy v. Duchess of Buckinghamshire*, reprinted annually for thirty years. Mountjoy 378, all English. Booyzen 0. Daschner 0. Booysen 6, all the Booysen Land and Mining Company. Control: the same feed and the same date range returns the five Kolbe notices.
93. Next — ready to start David
Find a FamilySearch affiliate library and read film 008004264
This is the shortest path left to the three missing sons, and it cannot be done from a desk. Film 008004264 holds the Savoy German Lutheran births and baptisms 1694–1840 in 537 images, padlocked at home and open at any FamilySearch Center or affiliate library — many public and university libraries are affiliates. Every volume has its own name index, so the job is: find Kolbe in the index, note the entries between 1808 and 1812 and 1814 and 1818, and photograph them. The same film also holds marriages to 1754 and burials to 1771, which reach back past this family into whoever the Kolbes were before Broad Street. Bring: the eleven known children with dates, and both spellings of the father — *Gottlob* and *Gottlieb*.
94. Done — finished and published this archive
Give each search in searched.json an outcome
This archive does hold its searched page as data — 104 dated sets — but as session write-ups (`g` headline, `n` note, `d` date, `h` link, four-column `rows`) with no outcome field on any of them. The shared register turns on exactly that field: it is how the page can say how many searches returned nothing, which is the difference between a register and a highlight reel. Writing outcome words from outside would be inventing findings, so this one belongs here. The word is yours to choose — the estate does not impose a shared five.
Raised by the UI alignment pass, 2026-09-16. It is recorded here because it is research this archive owns, not something to guess at from outside.
Done 20 September — and this row was wrong about its own archive. It said the 104 sets carry *no outcome field on any of them*. Every one of the 606 queries has carried one all along, as the fourth column of its row: yes 281 · no 157 · part 143 · blocked 13 · void 12. It was already there on 15 September, the day *before* this row was raised, and the page was already counting with it. The row had read the shape of a set — `g n d h rows` — and never opened a row. So there was nothing to write and something else to do. The page now carries the shared `Searched` register above the sessions: all 606 queries in one filterable list, with the archive's own five outcome words as the filter, beside the 112 session sections which stay because they are the narrative — what was asked in one sitting, and why. Both draw the same rows, so they cannot disagree. `nil` is `["no"]` alone: *blocked* means a paywall or a dead server and the search never ran, and counting thirteen unasked questions as thirteen empty answers would be a lie in the archive's favour; *void* is not a search at all. One kit change was needed — `Searched` built its haystack without the detail column, and of 3,436 distinctive terms in that column 2,933 appear nowhere else, so *Achilles* and *Aliwal North* were unfindable in a register of searches.
95. Done — finished and published this archive
Turn /open-questions/ into rows
270 lines of prose, the second longest in the estate. Shape: the question, what is known, what has been tried, and the record that would answer it. `kit/components/Questions.astro`.
Raised by the UI alignment pass, 2026-09-16. It is recorded here because it is research this archive owns, not something to guess at from outside.
Done 17 September. Extracted to `site/src/data/questions.json` and drawn by the kit's `Questions.astro`, which brings the filter bar and the four named fields. The extraction found a live fault rather than just moving text: the rows were tuples destructured as `[n, title, why, where, pri]`, and two of the sixteen are written with six fields — §0, this archive's oldest wall, and §12, James Montjoy's parents, both carry *what has been tried* between what is known and what would settle it. The destructure ran past it, so those two pages printed their whole what-would-settle-it paragraph in the slot where the priority chip goes, and captioned what had been tried “Where to look”. Both now render all four fields with the right labels. The eyebrow said Thirteen and there are sixteen; it is counted off the rows now.
96. Done — finished and published this archive
Extract /sources/ into data
674 lines written inline — the largest sources page in the estate. Only Falco and Defranceschi hold theirs as data, so the row cannot move past two of seven until the others do. Nothing about the content need change; it is the shape that blocks it.
Raised by the UI alignment pass, 2026-09-16. It is recorded here because it is research this archive owns, not something to guess at from outside.
Done 17 September. 674 lines → 548, with the four register sections — *documents read*, *references in hand*, *the Transvaal notices*, *the Transvaal estates* — lifted into `site/src/data/sources.json` by `tools/extractsources.py`: 42 rows and the 23 FamilySearch arks. The cells were carried across as the HTML they already were, so nothing is reworded and nothing summarised; the only change is that an internal link written as an Astro `u()` call became a plain site path, and the page puts the base back on at render. That is what makes it checkable, and it was checked: strip the tags off the rebuilt page and the reading text is byte-identical, 35,508 characters before and after, with the same thirty links in the same order. The rest of the page stays prose, because the rest of it argues rather than lists — the book that went the other way when it was read, the seven kinds of shut door, the library that is not an archive. One thing fell out of having rows: the Transvaal estates now carry a derived tally, and it says 10 references — 5 blocked, 3 read, 1 not found, 1 with no status recorded. That last one is TAB MHG O/2899, which states its position in prose and wears no chip. Three of seven archives hold their sources as data now, which is still short of a shared component.
97. Blocked — waiting on a person or an institution Free State Archives, Bloemfontein
VAB MHG M777 — the estate that would name the Commandant’s twelve children
This is now the most valuable unread document on the site. Commandant John Gerson Eliezar Kolbe, the Settler’s eldest child, died at Leeuwberg on 3 April 1868 leaving eight sons and four daughters, and this archive can put a hand on two. Today three more Kolbes appeared in that exact country — a man farming Leeuwberg in 1901, a Kolbe commando and a “General Kolbe” taken at Reddersburg in November 1901, and a G. A. Kolbe of Bethulie who was President of the Orange Free State Agricultural Association in 1914. None of them can be attached to anybody until the twelve children are named, and the document that names them is her 1896 estate, filed under her second husband’s surname: VAB MHG M777 — “MCCABE, ANNA JACOBA. NOOIENSVAN VAN DER MERWE. EGGENOTE JOHN GHERSON [KOLBE]”. Beside it sits VAB MHG K71, the Commandant’s own. VAB is Bloemfontein — a different building again from TAB, KAB and NAB, so it is a fourth trip and cannot go on the Pretoria researcher’s list. 20 SEPTEMBER — AND NOW IT HAS SOMETHING TO BE CHECKED AGAINST. A Geni profile for Frederik Fortunatus Kolbe (b. 22 Oct 1857 Bloemfontein, d. 18 Dec 1930) gives him as son of Commandant.Gen. John Gherson Eliezer Kolbe and Anna Jacoba, and names six of his siblings — Catharina Wilhelmina Magdalena Pienaar Anderson, George Augustus, General Willem Johannes, Jan Hendrik, Petrus Jacobus — *“and 6 others”*, which is twelve. It is a collaborative tree with one source attached and it is not evidence. But two of its claims match documents this archive reached independently: his wife Maria Petronella Elizabeth Antoinette and daughter Rosa Magdalene against the 1890 Bloemfontein baptism of Rosa Magdalena Kolbe to *Frederik Fortuna Kolbe and Maria Petronella Elizabeth Antoinetta McCabe*; and his daughter Maria Margaretha Downing Rousseau against TAB MHG 5450/68, an orphan index entry here. A tree says twelve names; this estate file proves them. 21 SEPTEMBER — NOW ORDERED WITH THE MEMOIR, IN ONE LETTER. M777 and K71 are asked for in `requests/free-state-archives-kolbe-three.md` alongside VAB A1 19, Renier-versameling No. 244, the *Herinneringe van genl. W J Kolbe* found the same day in a Pretorius footnote. Same building, one enquiry. The letter names the death notice as the page most wanted from each file, because that is the page that would name the twelve children. 27 SEPTEMBER — STILL WANTED, AND FOR LESS THAN IT WAS. Seven of the twelve now have fonts, found in the Free State and Cape registers on the mother's name. M777 would still name the four daughters and settle the sibship entire. AND THREE MORE ESTATES ARE NOW NAMED, ALL AT THE SAME REPOSITORY: Johan Casimir Kolbe, probate 1906 at Smithfield; John Henry Kolbe, probate 1924 at Ficksburg; George Augustus Kolbe, probate 1905 at Bloemfontein. None ordered. They are the Settler's son and two of the Commandant's, and they would be one enquiry rather than three.
98. Done — finished and published this archive
The 21st Light Dragoons at the Cape, read in the London papers 1806–1817
London Courier and Evening Gazette, 2 January 1812: *“CAPE OF GOOD HOPE. The 21st Light Dragoons are in the highest state of discipline. The following is a Copy of General Orders given [by] Sir John Cradock, after reviewing that regiment…”* — James Mountjoy’s regiment, at the Cape, reviewed by the Governor whose name the town he settled in carries. Only the first line has been read. The full order may name officers, give the regiment’s strength and station, and say what it was doing in 1812 — five years before Mountjoy married at Graaff-Reinet. There are 258 articles in the same search and none of the rest has been read either. ··· DONE, AND IT GAVE A DEADLINE. Sixty-seven articles read in date order. The regiment sailed from York for the Cape in May 1806, went on to South America and back, was reinforced in 1810 and by a draft of 140 men in October 1813, was reviewed by Sir John Cradock, and — in a letter from the Cape dated 1 and 10 September 1817, printed in five London papers — “THE 21ST LIGHT DRAGOONS HAVE EMBARKED FOR INDIA”. James Montjoy married at Graaff-Reinet on 9 March 1817, six months before. The table
99. Blocked — waiting on a person or an institution The National Archives, Kew
WO 12/1450 — the one piece that dates the discharge
WO 97 will not hold him and this archive has stopped expecting it to — the pre-1855 discharge papers cover only men discharged to a pension, and he took his papers at the Cape to marry and make saddles. WO 12, the muster books and pay lists, lists every man by name, quarter by quarter, including the quarter he leaves. The windows are now dated: he arrived with the regiment in 1806, with the reinforcement of October 1810, or with the draft of 140 men in October 1813 — and he was gone before the regiment embarked for India in September 1817. So the quarters to read are Michaelmas 1816 to Midsummer 1817 for the discharge, and the arrival quarters for the enlistment. WO 12 is not digitised for this regiment and this needs a reader at Kew. It belongs with the two Kew letters already drafted. ··· 17 SEPTEMBER — THE PIECE IS NAMED, AND THE OLD NAME WAS WRONG. Discovery gives the regiment a continuous WO 12 run: 1446 (1805–06), 1447 (1807–09), 1448 (1810–12), 1449 (1813–15), WO 12/1450 (1816–17), 1451 (1818), 1452 (1819–20). WO 12/1450 alone covers the window — held at Kew, no access conditions. The piece this archive had written into the drafted Kew letter, WO 12/1430, is the 20th and 21st Dragoons for 1867–1868 — fifty years late and the wrong lineage. The draft has been corrected and says so.
100. Done — finished and published this archive
Give the person page the evidence files it still cannot see
The 17 September audit found one live fault and one standing gap. The fault is fixed. The gap is that `/people/[slug]` reads only `dossiers.json`, `people.json`, `places.json` and `kin.js` — so everything else reaches a person only if it happens to be rendered as prose on some page and then picked up by the dossier gatherer. `marriages.json` is keyed to a husband and a wife by name and the person page never reads it; a marriage this archive holds shows on a person’s page only by that indirect route. `corrections.json` and `searched.json` are keyed to a page anchor rather than to a person, so a correction about a named person cannot be shown on their page at all. Done 17 September: the person page now reads `marriages.json` directly. Thirty-three pages gained a marriage record — the date, the town, whether a document or a family tree puts them there, the note and the source — where before the chart drew a spouse and the evidence for the marriage reached nobody. Matching is through `resolve()` in kin.js, on name and birth: Petrus Jacobus Booysen born 1788 and his son born 1812 share one page and have four marriages between them, and name-only matching would have handed the son's two to the father. `checkevidence.py` now refuses the build when a marriage it holds is missing from either partner's page, when the date is missing from the block, or when a reference resolves to more than one person; all three were proved by reintroducing them. What is still open is the larger half: `corrections.json` and `searched.json` key to a page anchor rather than to a person, so a correction about a named person still cannot be shown on their page. That is a schema change and it is row 102.
101. Next — ready to start the Geneology Map session
The worldwide source map keeps no work list, and everything it needs is already installed
Checked today at that project's own kit pin (`8174676`) and in its `node_modules`: `WorkList.astro`, `checkworklist.py` and `worklist_rollup.py` are all present and installed, so nothing has to be built or repinned to give it a list. What is missing is the wiring — no `site/src/data/worklist.json`, no `/worklist/` page, and a build script that runs `build-surnames.py && build.py && check-data.py` and none of the estate gates. Its work currently sits in two loose files, `notes/45-kit-promotion.md` and `requests/meertens-familienamenbank.md`, which nothing reads. The estate roll-up already counts it: 9 archives, 1,003 rows, 440 open, and two with no list at all — Our Family and the Geneology Map. Worth doing at the same time as a repin, because that project is 52 commits behind the kit. The rows are that session's to write; this archive is not going to invent work on its behalf. The work list
102. Done — finished and published this archive
Corrections and searches key to a page, not to a person, so neither can be shown on a person's page
The other half of row 100. `corrections.json` rows are positional — date, was, now, cause, href — and the `href` is a page anchor: documents 35, open questions 4, name 4. `searched.json` is the same shape, a dated session write-up attached to a page. So a correction that is entirely about one named person can be read on /corrections/ and cannot be read on that person's page, which is where somebody looking them up will be. This is a schema change and it touches two files and every page that renders them, which is why it is separate from row 100 rather than folded into it. It is also the thing the 17 September audit called the standing gap, as against the fault, which is fixed. The corrections
Done 17 September, the corrections half. `tools/correctionswho.py` works out which people a correction is about and 44 person pages now carry theirs — 31 of the 63 corrections name somebody this archive holds, 58 attributions in all. A person is attached only where their whole held name appears, forename and surname together, never a surname alone: matching on part of a name is how this estate once turned Domenica Prostamo into Domenica Anile. Where two held names overlap the longer one wins, so *George Augustus Kolbe of Wakkerstroom* is not also filed as *George Augustus Kolbe*, who is a different man kept deliberately apart. Every attribution is printed for review and one can be refused in `corrections-who-deny.json`. The derivation is gated — `check:corrections` refuses the build when the written file is not what the tool would derive today, proved by dropping an attribution and by pointing one at the wrong person; both exit 1. `searched.json` is still page-keyed and is the other half; it is a session write-up rather than a claim about a person, so whether it belongs on a person page at all is a question, not an oversight.
103. Next — ready to start the D'arcy session
D'Arcy's work list says two archives keep none, and names the wrong two
Raised here because it belongs to another archive and would otherwise go unsaid — which is exactly what that row itself says about the archives it names. The D'Arcy row reports the roll-up as 8 archives, 153 rows, 113 open, and two with no list at all: Luwinski and Our Family. Luwinski has kept one since then. Run today the roll-up gives 9 archives, 1,003 rows, 440 open, and the two without a list are Our Family and the Geneology Map. The row is still true in its point and wrong in its figures, which is the failure a roll-up built from the archives was written to end: the numbers were typed in rather than read out. The work list
104. Done — finished and published this archive
The priority on an open question is doing two jobs, and it is stale on seven of the sixteen
Raised by the extraction in row 95. The `state` field on a question holds Blocking, High, Medium, Low, Answered, Closed — a priority and a state of play mixed in one column. Only two rows have ever been moved to Answered or Closed, while seven more announce an answer in their own heading: §3 “SEARCHED, AND NOT FOUND”, §5 “ANSWERED. IRELAND.”, §6 “BROKEN”, §8 “DONE.”, §9 “CONFIRMED”, §10 “YES. ONE DID”, §13 “ANSWERED, 16 SEPTEMBER 2026” — and all seven still read High, Medium or Low. The page will not derive it from the heading, because reading a state out of a sentence instead of a field is how the D'Arcy register came to show twenty-one medieval manuscripts as “family tree”. So the eyebrow now reports the tally of the field and claims nothing about how many are open. Which of the seven are shut, and whether a question that has been answered keeps a priority at all, is a judgement this archive has to make. The questions
Done 17 September, one at a time and by hand. Five of the seven are shut and two are not, which is the whole reason this was not derived. §5 was answered by the 1903 death registration entering him as IRISHMAN BORN IN IRELAND; §6 by Yorkshire burials returning zero Barrys at Bilsdale inside an indexed run; §8 by there being nothing left on those profiles; §9 by the settler roll; §13 by a relationship search naming Margrith Louisa Maria (1852) and the second missing child. §3 and §10 stay open at the priority they had. §3 reports a negative result from one database and names two unread sources that would settle it — the Cape camp registers proper, and the Cape refugee and ‘undesirables’ returns. §10 corrects a claim this archive had made rather than closing the question, and still points at mining company registers, Johannesburg street directories, five Transvaal estate files and the Indwe registers. A rule reading the headings would have shut all seven and been wrong about two of them. The page now carries the reasoning, and the eyebrow reads *Sixteen — 1 blocking, 5 high, 3 medium, 6 answered, 1 closed*.
105. Done — finished and published this archive
The Burgersdorp baptism sweep — finished, 605 to 818
Film 008039088, the Burgersdorp Dutch Reformed baptism register, 22,059 records. The hand sweep stopped at image 604 of 818 when the FamilySearch session expired, and it has been stopped ever since. 214 images unread, and this is the register that has already produced the 1851 baptism, the 1889 font with both maternal grandparents standing, and the Mountjoy entry three weeks later on the facing page. Two traps already recorded and both still apply: the personas API wraps silently at offset 1000 and will re-serve page one while reporting 22,059 results, so the sweep must be by image and not by offset; and `q.filmNumber` is scored, not filtered, when combined with any other term. ··· 17 SEPTEMBER — BLOCKED AGAIN, ON THE SAME THING. The film viewer redirected to `ident.familysearch.org` asking for a username and password. This is the second time the session has expired mid-sweep, and it is why the sweep stopped at 604 in the first place. The pattern is now worth naming: a 214-image hand sweep is longer than a FamilySearch session lasts, so it has to be done in sittings with the image number written down at the end of each, which is what the `604` in this row is for. Nothing else about the job has changed and no credentials were entered. ··· DONE THE SAME DAY, AND NOT BY EYE. The viewer’s Image Index panel is fed by `POST /search/filmdatainfo/image-data`, which takes an image ARK and returns every indexed record on that image with both parents, both dates, the place and a citable reference. 214 images, 2,046 baptisms, 1858–1866, no image skipped. Five rows were ours. The sweep
106. Done — finished and published this archive
The three unfollowed Kolbe children, and Angeline of Wakkerstroom — all four answered
Also parked behind the FamilySearch sign-in since 15 September. Three of the Settler’s Colesberg grandchildren are not followed out of the font, and an ANGELINE KOLBE dies at Wakkerstroom in 1900 while a MARTHA ANGELINE KOLBE is “buried at Volksrust” in 1901 aged two months — withdrawn on 14 September 2026: her stone is in WAKKERSTROOM MAIN CEMETERY, and the index gave the district rather than the grave. — neither attached to the Martha Angelina baptised at Colesberg in 1868, and this archive will not guess. Wakkerstroom has since become the town where three separate threads meet: C.K. Barry the attorney, the George Augustus Kolbe of the gravestones, and the 1901 prisoner. The Angeline question is now worth more than it was when it was parked. ··· DONE, AND BY ONE SEARCH. The Pietermaritzburg estate files hold 64 Kolbe entries. Both parents’ estates — hers 1923, his 1929 — name nine children, the daughters under both surnames. Gabriel Stephanus and Martha Angelina are alive at both; Catharina Maria Louisa married a LOUW; and because Martha Angelina was alive in 1929 she is neither the Angeline who died at Wakkerstroom in 1900 nor the Volksrust infant of 1901. The question is answered by exclusion. Also: the parents’ deaths — Utrecht 1929 and 1923 — and a ninth child. The two estates
107. Done — finished and published this archive
The phthisis reading on Henry James Booyzen — and both causes were wrong
The fourth job parked behind the sign-in. A cause of death read once and never checked against the register image. A disease named from a transcript and not from the document is exactly the class of citation this archive struck on 16 September when R_177176049779 turned out not to resolve. ··· DONE. THE IMAGE IS OPEN AND IT SETTLES IT BETTER THAN EITHER VERSION. Phthisis is on line 11 — “Pensioner or Recipient of Maintenance Allowance” — not line 12. It is what he was paid for, as a certified miners’ phthisis pensioner, not what killed him. Line 12 reads “Cor pulmonale with Silicosis”, contributing cause broncho pneumonia. And the burial register’s “coronary pulmonary sclerosis” is a mangling of that same line. The form
108. Next — ready to start the Falco, D'arcy and Luwinski sessions
Three archives have spelling clusters waiting on their own session's reading
Part of the estate-wide search folding done on 17 September. `namefold.py` folds a spelling on sight when it is one word two keyboards wrote differently — an accent, a capital — and refuses to fold anything that changes a letter, because whether two spellings are one family is an archive's judgement and not a rule's. Five archives had safe clusters and now fold. These three have none at all, and between them sixteen forms are waiting: Falco Sansone → Sanzone (312 tokens, and typing Sanzone currently reaches 15% of that name); D'Arcy Sanyger → Saniger (22), ZANIGAR → Sanigar, Lion → Lyon; Luwinski LUWINSKY → Luwinski (18), Hostowski → HOSTOWSKY. Run `namefold.py --root site --all` once the session has read them. Also waiting: Blazevic Blasevic → Blažević (52) and Mazza Ericksen → Eriksen (8).
109. Done — finished and published this archive
Four archives keep their name pages in four different menu groups
Raised by David on 17 September, looking at two sites side by side: Blažević files Say Your Name under Origins and Defranceschi files it under The name. Measured across the estate, every archive shares a spine — Start, Our line, The families, People & places, Evidence — and beyond it the name pages sit in five different parents: *The name* (Defranceschi, Falco, Mazza), *Our line* (Blažević, Booyzen, D'Arcy, Lerena), *Origins* (Blažević), *People & places* (Booyzen's forenames), *People & names* (Luwinski). Blažević's Origins group is really about where the family came from — the Bunjevci, the Frontier, the Map, Port Pirie — and the two name pages are lodged there because there was nowhere else to put them.
Done 17 September. The canon is taken from the three archives that already had it rather than invented: where an archive has two or more name pages they sit in a group called “The name”, between People & places and Evidence. This archive now has one (The Name, The forenames) and Blažević has one (The Name, The Second Name, Say Your Name) — which also gives Origins back to origins: the Bunjevci, the Frontier, what they did, the other emigration, the map, Port Pirie. Two documented exceptions. *Lerena* keeps one name page and keeps it under Our line, because a rule that forces a group of one is a rule making work. *D'Arcy* keeps Fifteen Spellings in The Sanigers, where it is the seventh of ten pages in a narrative about that name; lifting it into a group of two would strand it from the nine pages that give it its point. That one belongs to the D'Arcy session to confirm or overturn.
110. Next — ready to start the Luwinski session
Luwinski's menu spine differs from the other seven and nobody has said whether that is deliberate
Found while settling the name group on 17 September. Seven archives run Start · Our line · The families · People & places · Evidence. Luwinski runs Start · Our line · Places · People & names · Evidence — no families group, places split out on its own, and *People & places* renamed *People & names* to house a single page, Her names. It may be right: this archive is one woman and her documents rather than a family, so a families group would be empty and the names question is the whole subject. But it has never been said out loud, and a deviation nobody has written down is indistinguishable from drift.
111. Next — ready to start the Defranceschi session
Defranceschi's name forms now exist in two files, and only one of them has a gate
Left behind by the shared Say Your Name component on 17 September. `givennames.json` is still the source of truth and still what `name-forms-audit.py` checks; `givennames-forms.json` is the shared shape the page now renders, converted mechanically from it. The conversion was checked exactly once — 645 forms, 419 attested, 226 aids, all three matching the gate — and nothing will check it again. Either the audit should read the shared shape and the old file go, or the build should refuse when the two disagree. Two files holding one fact is how a number goes stale, and this estate has been caught by that more than once this week.
112. Next — ready to start the six other sessions
Six archives could adopt Say Your Name and none of them holds the data yet
The component is in the kit and two archives are on it. What the other six need is the data: `{canon, langs: {<key>: [{form, seen, note, case}]}}` plus their own list of languages. The material exists in every one of them — this archive has eleven Afrikaans/English doublets running to 8,931 occurrences, D'Arcy has fifteen spellings across 340 entries, Luwinski already holds `forms:[{form, kind, where, note}]` per person which is the shape in all but name, Falco and Mazza have Latin and Italian against each other in every register they read. Writing the table is each archive's own work and not something to guess at from outside; the page that draws it is no longer an obstacle.
113. Blocked — waiting on a person or an institution Pietermaritzburg Archives Repository
Order the two Pietermaritzburg estate files — MSCE 1923 and 1929, Kolbe of Utrecht
The index gave nine children, two deaths and two birthplaces. The files themselves would give more and none has been read. Elizabeth Johanna du Toit Kolbe, died 12 July 1923 at Saxony, Utrecht; Frederick Fortunatus Kolbe, died 23 November 1929 at Utrecht. A Natal estate file carries a death notice — which names the deceased’s own parents, and would say whether Frederik Fortunatus’s father is written as the 1820 Settler. That single line is what the whole Kolbe descent still rests on inference for. The files also give husbands’ forenames for the five married daughters, of whom Louw is currently a surname and nothing else. This is a fourth archive building — NAB, not TAB, KAB or VAB — and it is the same repository the Pietermaritzburg letter has been drafted and unsent for since 12 September. The letter should now carry these two references. There are 64 Kolbe entries in that collection and this list names two.
114. Done — finished and published this archive
Read the DSAB entry on George Augustus Kolbe, and work its source list
Melanie Geustyn sent a second document nobody asked for: two pages written for the Dictionary of South African Biography by Dr Karel Schoeman, who worked at the National Library, found in the MSB 751 Background File, with a list of Bronne (sources). It is a secondary work and this archive does not take secondary works as evidence — but a source list by a historian of Schoeman’s standing is a reading list, and it is the only one this archive has ever been handed for this man. Read it, and then work every source in it that is not already here. DONE 20 September 2026. Read in full and set against the archive line by line. It agrees on nine points reached from nine different documents and is wrong about exactly one thing — it makes the Settler’s father a German apothecary, where 121 rate books, two printed masters’ lists, three Gazette notices, the census, the burial register and a sworn Old Bailey deposition all say tailor. The trade is the son’s, moved one generation up. It also corrected this archive (MSB 751 is not a letter “on his ordination”), settled the farm name (Wurthemberg = Knapdaar), and handed over nine years of his life this archive did not have — including an accusation, an inquiry and an acquittal in 1836. Its eleven sources are now rows of their own.
115. Done — finished and published David
Thank Melanie Geustyn, and tell her what the letter produced
She answered in twenty-three hours, sent the manuscript on the third day, chased the Background File unasked, and added a DSAB entry nobody had asked for. That is the best service any institution has given this archive. The reply should say what the letter actually did: it did not carry the address, and it settled the LMS question, dated the obituary, gave Henry his illness, gave Elizabeth a nickname and a city, and turned Charlotte from an inference into a daughter. It is also worth telling her that the Background File negative was useful — archivists rarely hear that. The draft is this archive’s; the sending is David’s. DRAFTED 20 September 2026 — requests/nlsa-thank-you-geustyn.md. The sending is David’s. It goes as a reply on her own thread, tells her concretely what the four pages settled and what they did not, says that the second attachment corrected this archive, and carries exactly one question, marked optional: the DSAB typescript is unsigned, and the Library is the only body that can say who compiled it and whether it was ever published.
116. Next — ready to start this archive
The Charles Bell picture of Kolbe at Philippolis, 1834 — found in Edinburgh, original in Johannesburg
FOUND, 20 September 2026, and it is in two countries. The DSAB sent this archive to the Africana Museum in Johannesburg (now MuseumAfrica, Newtown) for a watercolour by C. Bell of a church service at Philippolis in 1834 with Kolbe in the pulpit, reproduced in *The Journal of Andrew Smith*, ed. Lye (Balkema, 1975). It is also catalogued in Scotland. University of Edinburgh, Centre for the Study of World Christianity, CSWC 60 — five photographs, 1834–1840, *“photograph copies of sketches by the South African artist C. Bell of African congregations”*. Item 3 is “Mr. Kolbe”, “at Philippolis in 1834”. The other four are James Archbell, William Edwards, Jean-Pierre Pellissier — who is source 9 on the DSAB’s own list — and a service at Ebenezer. NEXT: ask Edinburgh for a copy of item 3, and MuseumAfrica for the original. Edinburgh holds a surrogate and is a university archive, which is usually the easier of the two. This archive holds no likeness of anybody born before 1850. 21 SEPTEMBER — MUSEUMAFRICA HAS NO ONLINE CATALOGUE TO SEARCH. Confirmed that Bell’s originals are there: he made some 300 watercolours and drawings on Andrew Smith’s expedition and a large number of the originals are in the Africana Museum, now MuseumAfrica in Newtown. But there is no searchable online collection catalogue, so the specific Philippolis 1834 sheet cannot be confirmed from here. The museum published printed catalogues of its Africana, which is the other route. Contact: `museumafrica@joburg.org.za`. The Edinburgh letter remains the better first move — CSWC 60 item 3 is catalogued, identified and in a university archive that answers email. 21 SEPTEMBER — THE MUSEUMAFRICA LETTER IS NOW DRAFTED, at `requests/museumafrica-bell-philippolis-1834.md`, so both halves of this row are written and both are David's to send. It asks three things in order of usefulness — the accession number, a reproduction and its fee, and where the sheet went if it is not theirs — and it offers the museum the other four names in the series (Archbell, Edwards, Pellissier, Ebenezer), because a collection with no online catalogue may file the group under a name that is not Kolbe's. The Edinburgh letter is still the better first move and the two should not be sent as a pair: CSWC 60 item 3 is catalogued, identified, and in a university archive that answers email, and if Edinburgh sends the surrogate it will carry details that make the Johannesburg request sharper.
117. Next — ready to start this archive
Burger Olivier, *So onthou ons die oewerbewoners van ou “Grootrivier”* (Bloemfontein, 1973) — it carries a portrait of him as a young man
The DSAB entry says a portrait of Kolbe as a young man was in 1992 held by a descendant, C.J. du Plessis of Ruigtevlei, district Bethulie, and that it is reproduced in this book. A privately published Free State local history of 1973 — the kind of book that sits in one or two libraries and nowhere else. NLSA, the Free State provincial library and Worldcat are the places to look. If the plate exists, the second question is whether the du Plessis family still has the original. 20 September — the book is identified and no copy is for sale. Printed by N.G. Sendingpers, Bloemfontein; a South African secondhand dealer has listed it, and searches for an available copy return nothing. The realistic holder is the National Library of South Africa by legal deposit. Do NOT bolt this onto the thank-you to Melanie Geustyn — that letter already carries two optional questions and a third would be pushing an institution that has been generous. Ask separately, after she replies. The deeper route is the other one: the portrait itself was in 1992 with C.J. du Plessis of Ruigtevlei, district Bethulie, and a family may still have the original. 21 SEPTEMBER, SECOND ATTEMPT, AND NO CATALOGUE REACHES IT. Searched again for holdings: nothing in WorldCat results, nothing in any library catalogue that surfaces, and no copy for sale. The publisher is confirmed as N.G. Sendingpers, Bloemfontein — a Dutch Reformed Church press — which is consistent with a locally printed 1973 local history that was never widely distributed. This is not going to be solved from a desk. The realistic holders are the National Library of South Africa by legal deposit and the Free State provincial library, and the route is a letter. Not to be bolted onto the Geustyn thank-you, which already carries two optional questions.
118. Struck — founded on a mistake, kept rather than deleted Western Cape Archives
Order the Settler’s death notice — STRUCK: it was ordered and read before this row was written
STRUCK THE SAME DAY IT WAS RAISED. This row was created from the DSAB source list on the assumption that MOOC 6/9/35 no. 7698 was a reference this archive did not have. It has been read since before September. It is on /documents/ with two photographs and a full transcription: born London, parents left blank, aged 42, *Agriculturist*, died 1 December 1844 “at Knapdaar, in the field cornetcy of Rhenosterberg”, property about £1000, signed M. Kolbe by his widow at Colesberg in January 1845 — and it names ten children. The row was raised because the DSAB write-up was built from the source list without first checking the source list against this archive. That is the same fault this archive logged on 17 September when a sweep’s results were called new before being checked against what was already held, and the rule written then — *a sweep that returns a name is checked against the archive before it is called a find* — applies to a bibliography exactly as it applies to a name index.
119. Next — ready to start David
Albany Museum, Grahamstown — the photocopy of the 1819 Kolbe–Downing marriage certificate
This archive holds the allegation — the sworn statement made to obtain the licence, and the place where George Augustus’s age was sworn falsely. It does not hold the marriage. The DSAB source list says the Albany Museum in Grahamstown has held a photocopy of the certificate since before February 1992. A certificate gives the parish, the witnesses and the signatures, and Margaret signed the allegation with a mark. Write to the Albany Museum. DRAFTED 20 September 2026 — requests/albany-museum-kolbe-marriage-1819.md. The sending is David’s. To `info@am.org.za`. The Albany Museum keeps a genealogical archive of the 1820 British Settlers, which is precisely what he was, so the letter asks a second question as well: whether anything else is filed under Kolbe or under Owen’s party. What this archive has of his first four years in the Cape is one sentence in a secondary work.
120. Next — ready to start David
Schoeman (ed.), *Griqua Records: the Philippolis Captaincy, 1825–1861* — to test the 1836 accusation
Van Riebeeck Society, Second Series no. 25 — 162 official and semi-official documents of the Philippolis Captaincy, including the Philippolis Law Book and the correspondence between the Captaincy’s officials and the missionaries. H. Hendricks, who led the agitation against Kolbe in 1836, was secretary of the Griqua Council — his papers are exactly what this volume prints. It was published two years after the DSAB draft was typed, which is why its own editor could not cite it there. Sold as an eBook by HIPSA (hipsa.org.za); no free digitisation found. This is the one purchase on the whole work list that would test a specific published allegation against a member of this family.
121. Next — ready to start David
Buy Schoeman (ed.), *Early White Travellers in the Transgariep, 1819–1840* — it prints the 1828 journey
The second document George Augustus Kolbe wrote himself, and the record of the November 1828 journey to the Caledon valley that the DSAB calls the earliest in the present Free State to survive in writing. Searched the Internet Archive by title and Google Books; not found. The Internet Archive holds no run of the LMS *Quarterly Chronicle of Transactions* and the Books API rate-limited the query. Next attempts: the Google Books interface directly, HathiTrust (whose bot check blocked this archive once already), and the CWM archive at SOAS, which holds the manuscript journals behind the printed extract and is source 4 on the same list. 20 SEPTEMBER, UNBLOCKED. The search under KOLBE failed because the printed version is filed under the other man’s name. The full citation is “Extracts of the Journal of Messrs Melvill and Kolbe, addressed to the Rev. Richard Miles, Superintendent of the Society’s Missions in Africa, Philippolis, 25 Nov. 1828”, Council for World Missions (LMS) Archives — which is source 4 on the DSAB’s own list, two lines above source 6, and the entry never says they are the same document. “An edited version, dated 18 Nov. 1828, appears in Melvill (1829).” AND THE JOURNEY IS PRINTED IN FULL, IN ENGLISH, IN A BOOK STILL IN THE SHOPS: Karel Schoeman (ed.), *Early White Travellers in the Transgariep, 1819–1840* (Pretoria: Protea Book House, 2003), ISBN 978-1-86919-016-3, page 30 — “The journey of John Melvill and G.A. Kolbe from Philippolis to the Caledon (1828)”. About ZAR 285. It is an edited collection of the travellers’ own texts, so it should carry his account entire. That is the fourth Schoeman title standing behind one unsigned typescript.
122. Done — finished and published this archive
Sibree, *A register of missionaries* (LMS, 1923) — found, free, and it holds his whole mission life
One printed line would carry his LMS service dates, and this archive now has three to test: mission assistant 8 February 1828, ordained 25 May 1833, dismissed 19 June 1837. A register of this kind also records why a man left the Society. Not found on the Internet Archive under that title; try Google Books, HathiTrust and the SOAS library catalogue. DONE 20 September. Free PDF, 225 pages, at missiology.org.uk — this archive had searched the Internet Archive, failed, and written it up as *not found*, which was a negative about one shelf reported as a negative about the world. Entry 278, KOLBE, George Augustus, gives the whole of his LMS service in eight lines and settles Mr Miles: *“accepted provisionally by MR. MILES, Feb. 8, 1829, which action was confirmed by the Directors, July 29, 1829”*. It also disagrees with the DSAB draft on the year of acceptance, the day of the ordination and the men who performed it. And entry 522 makes F.W. Kolbe the Settler’s son, where Schoeman says he was not related.
123. Done — finished and published this archive
The 1820 settler roll of the *Nautilus* — read, and both ages are false again
The DSAB says the roll gives Kolbe’s occupation as apothecary and his age as 20. He was seventeen. That is the second age this man is known to have overstated — he had already sworn a false age to marry at sixteen the year before. The roll would also list who else was in Owen’s party, which is the company he landed with and a body of people this archive has never looked at. DONE 20 September. Nash’s Settler Handbook entry for Owen’s party is published in full at 1820settlers.com. “KOLBE, George Augustus 20. Chemist. w Margaret 19.” He was sixteen, sailing four days before his seventeenth birthday; she was sixteen. And it does not agree with the marriage allegation eleven months earlier, where both swore to “twenty-one years and upwards” — nobody twenty-one in January is twenty in December. They were not keeping a story straight; they were telling each paper what it needed. The party was 52 people, No. 3 on the Colonial Department list, joint-stock, recruited in London by Thomas Owen of 9 Carey Street, Horseferry Road, Westminster, whose father the Rev J Owen of Fulham handled the correspondence — and it collapsed in 1821 when Owen refused to draw rations and left.
124. Blocked — waiting on a person or an institution David
Chase the National Archives Repository, Pretoria — sent 13 September, nothing back
Sent 13 September 2026 to `enquiries2@dsac.gov.za`, copied to `Nontembekom@dsac.gov.za`, asking for the Transvaal estate files. No bounce and no reply in a week. The file itself records the distinction this archive learned from the SANDF enquiry: *no bounce is not the same as delivered.* AND IT HAS BEEN OVERTAKEN. A private researcher, Anne Brest, was engaged on 16 September and goes to Pretoria on Tuesday 22 September for the same seven files. So the question this row now asks is not “when will they reply” but whether to withdraw the request once the photographs are in hand, so that staff time is not spent on files this archive already has. 20 September — see `notes/anne-brest-tuesday-addendum.md`.
125. Blocked — waiting on a person or an institution David
The WikiTree message to Ewald Schmidt is typed and cannot be sent from here
Composed in the form on 13 September and stopped at a human-verification question — *“How do you spell the word for the number 10 in English?”*, labelled anti-spam protection. That is bot detection, and this archive does not go around bot detection; the same rule stopped it at HathiTrust and the reason is recorded both times. The message concerns estate MHG 10/57 and it is ready, word for word, in `requests/wikitree-ewald-schmidt-booyzen-10-57.md`. It needs a human to type one word.
126. Blocked — waiting on a person or an institution David
Chase the SANDF Documentation Centre — one address delivered, no reply since 9 September
Two addresses were tried. `sandfdoc@mweb.co.za` bounced — *550 5.7.1 Relay access denied* — and has never been replaced. `archive@dod.mil.za` was delivered and has been silent since 9 September 2026. The enquiry is for the pension file. This is the letter that taught this archive to check a mailbox after sending and to write down which half arrived, which is now a rule applied to every letter it sends.
127. Next — ready to start David
Write to Edinburgh — CSWC 60 item 3 calls Kolbe “a moravian missionary?” and he was LMS
A correction this archive can make to somebody else’s catalogue, and should. The University of Edinburgh’s description of CSWC 60 item 3 reads “Mr. Kolbe, a moravian missionary?, at Philippolis in 1834” — with the question mark, so the cataloguer already doubted it. He was London Missionary Society. Two independent documents say so: the catalogue of MSB 751 at the National Library of South Africa (“George Augustus Kolbe, 1802–44. LMS missionary”) and his father’s letter of 23 August 1830, which reports him “fully accepted by the [L.M.S.] as one of their Missionaries”. The DSAB draft adds the dates: mission assistant 8 February 1828, ordained 25 May 1833 by Arbousset and Casalis, dismissed 19 June 1837. The same letter should ask for a copy of item 3. One email does both. DRAFTED 20 September 2026 — requests/edinburgh-cswc60-bell-kolbe.md. The sending is David’s. To `heritagecollections@ed.ac.uk`, Centre for Research Collections, George Square. It offers the correction with both sources, tells Edinburgh that the original watercolour appears to be in Johannesburg — neither catalogue mentions the other — and asks for a reproduction of item 3 for private family-history use. It is the first letter this archive has written that gives an institution something before it asks for anything.
128. Struck — founded on a mistake, kept rather than deleted this archive
Read whatever Anne Brest brings back from Pretoria — STRUCK: the visit was stood down
Tuesday 22 September 2026. Seven files on Invoice No 463 — six Booyzen/Daschner estates and the Wear file she is giving for nothing. The death notice is the page that matters in each: father, mother, birthplace, occupation, surviving spouse and the list of children. TAB MHG 52653 is expected to fail and David has twice told her not to hunt it; if it is not there the useful outcome is the correct reference from the paper index, which she is the one person able to consult. Write up the misses at the same length as the finds — a “not there” from inside the building, with a reason, is a controlled negative. STRUCK 20 September. The Tuesday visit was cancelled two days out and Invoice 463 set aside. Nothing is coming back from Pretoria this week. This row returns when a settled list has been priced and paid. 22 SEPTEMBER — 52653 IS A FILE AND THIS WAS WRONG. Anne Brest had located it on NAAIRS on 15 September and said so, with a link: the absent year suffix is normal for some 1920s references. She was told twice not to hunt it after she had already found it. The correction.
129. Done — finished and published this archive
“Mr Miles” identified — he is the officer who accepted Kolbe, not merely the man who carried the news
An inference this archive is publishing as an inference, and would rather have as a document. The letter of 23 August 1830 says only *“Mr Miles had called at our house”*. The journal of 25 November 1828 is addressed to the Rev. RICHARD MILES, Superintendent of the Society’s Missions in Africa, and Richard Miles — formerly pastor of Brigg in Lincolnshire, originally appointed to Demerara — was sent to Cape Town to act as Dr Philip’s substitute while Philip was in England. The office, the continent and the dates all fit: Philip in England from 1826, Miles acting at the Cape in 1828, Philip back and Miles in London by 1830. What would settle it is the LMS’s own records of its officers — Sibree’s register, or the CWM home correspondence at SOAS, both already on this list. If it holds, the man Kolbe reported to from Philippolis is the man who knocked on his father’s door. SETTLED 20 September, by Sibree’s register. *“Accepted provisionally by MR. MILES, Feb. 8, 1829, which action was confirmed by the Directors, July 29, 1829.”* Miles did not carry the news — Miles made the decision. And the father’s adverb turns out to be the Society’s own vocabulary: provisionally by Miles at the Cape, confirmed by the Directors in London, and a father told his son had been “fully accepted”. Richard Miles was Acting Superintendent of the LMS at Cape Town from 1825, standing in for Dr Philip while Philip was in Britain. The letter still gives no forename, so the surname is carried by the office and the dates — but the office is now his own.
130. Done — finished and published David
Anne Brest — SENT: stand down from Tuesday, invoice aside, a checked list to follow
URGENT, AND IT IS A PROMISE NOT KEPT RATHER THAN A DISPUTE. David wrote on 16 September: *“please send the invoice and I will pay it straight away”*. Invoice No 463 is R11,400 plus R200 and cannot be funded before Tuesday 22 September, which is when she drives to Pretoria and for which she has already arranged the files. The one wrong move is silence. She volunteered a cap on her own fees before anyone asked, and gave a seventh file free. Drafted as `requests/anne-brest-payment-and-scope.md`; the sending and the choice are David’s. It says plainly that the payment will not arrive in time, apologises for having said otherwise, and puts the decision in her hands: move the visit to a date after payment, or keep Tuesday with a reduced set — Daschner MHG 3734/46, Booyzen MHG 1095/48, and Kolbe MHG 40293 substituted in, not added. It also withdraws the whole of Groups A, B and C from the 16 September email, so that nothing is being worked that is not on a paid invoice. Her fees are not the problem and the letter says so. R2,850 for the first file including the round trip to Pretoria, then by the work done and capped, is a professional day’s rate. The cost grew because the scope grew — seven files became twenty-six items in a single reply — and that is this archive’s doing, not hers. AND THE TIMING WAS NEVER POSSIBLE ANYWAY. Today is Sunday 20 September and the visit is Tuesday the 22nd. Her invoice states that Remitly, Wise and PayPal are not accepted — it is a SWIFT transfer to First National Bank, and an international transfer from Australia sent on Monday does not land in a South African account on Tuesday; two to five business days is normal. The visit was never going to be prepaid on this timeline whatever the balance said. The draft now leads with that, because it is true, it is checkable, and it is nobody’s fault — and it leaves the genuine part of the apology, which is the scope, standing on its own. 20 SEPTEMBER, CORRECTED BY DAVID. The draft first put MHG 3734/46 (Daschner) at the head of the reduced set “for her parents” — which is the reason the original 15 September letter gave. Her parents are already documented here, from her baptism at Barkly Oost, 24 May 1891, entry 120, and the same is true across that generation: Henry James Booyzen’s 1887 Barkly East baptism names Willem Hermanus Booÿsen and Catharina Maria Sophia Barry outright. The reason had gone and the letter still carried it, because the list was worked without being checked against the archive — the third time today. Re-ranked on what is actually missing, which is CHILDREN, not parents: MHG 40293 (Kolbe, a different question entirely), MHG 769/52 (John Barry — his stone says *aan ons vader* and not one of his children is named here), MHG 5155/51 (Gert Jacobus, from the second marriage, the thinnest branch). SENT 20 September 2026 (Gmail `1a0bc8e25e24070d`), on David’s instruction and as a reply on her thread. The two-option letter was abandoned before sending. David’s judgement was that putting another decision in front of her was the wrong move when the list had already moved four times, and he was right: the churn was the problem, not the money. So it asks her to stand down, sets Invoice 463 aside, promises no date, and offers to cover what the short notice has cost her.
131. Done — finished and published this archive
Dr Philip’s promised letter — folded into the SOAS request, which names it
“Dr Philip was exceedingly glad to hear of you being accepted by the L.M.S. as you may well imagine, HE HAS PROMISED TO WRITE TO YOU A LETTER BY THE FIRST OPPORTUNITY.” Written by the father in London on 23 August 1830. If Philip wrote it, it is another document with this man’s name on it, and it would be filed where the rest of the Society’s South African correspondence is — the CWM archive at SOAS, which is already on this list twice over, for the 1828 journal and for the incoming letters. One request could ask for all three. 20 September — folded into the SOAS letter, which now carries exact references for the journals and the correspondence years. Philip’s promised letter would sit in the same incoming-correspondence series if it survives. CLOSED 21 September: it is item four in `requests/soas-cwm-kolbe-journals.md`, which asks the Council for World Mission archive for the journals, the 1835–37 incoming correspondence and this letter together. Nothing further can be done here until that letter is sent.
132. Done — finished and published this archive
The Mirror story — FOUND: it is fiction, by James Hogg, and the tailor was right not to believe it
The father devotes two of his four pages to it and asks his son to say whether it is true. *The Mirror of Literature, Amusement, and Instruction* ran weekly from 1822 and is digitised in several places. The story: a Scottish colonist near the Xhosa, orang-outangs, a shot young animal that proves to be a chief’s son, a wife carried off, Capt Johnson and a company of soldiers, and a recovery after nearly two years. It would date to 1829 or the first half of 1830. Worth finding for its own sake: it is the only thing this archive has that shows what the London household was reading about Africa while its son was living there. 20 September — SEARCHED AND NOT FOUND. Google Books tried on “Melvill and Kolbe”-style phrasing and on the story’s own terms (“ourang outang” / “orang outang” with *carried off*, *wife*, *colonist*, *Caffre*, *Mirror*, 1829–1830) — no Books matches at all — and the same on the open web. Recorded so nobody repeats it. Next attempts: the runs of *The Mirror* on Project Gutenberg and the Internet Archive, searched by volume rather than by phrase, and the distinctive surname Williams or Williamson with Capt Johnson. FOUND 21 September, after five failed searches, by downloading the volumes and reading them. *The Mirror of Literature, Amusement, and Instruction*, Volume XIV (1829), pp. 313–318: “SPIRIT OF THE PUBLIC JOURNALS — A SINGULAR LETTER FROM SOUTHERN AFRICA. Communicated by Mr. Hogg, the Ettrick Shepherd.” It is JAMES HOGG, author of *The Private Memoirs and Confessions of a Justified Sinner*, printed as though it were a settler’s letter home. Every detail in the 1830 letter is in it: the chief KAROO offering oxen and “a choice virgin to boot” for the settler’s wife; “he foretold I should repent it”; captain Johnstone and nine British soldiers; the wife “carried across the river by a band of pongos, (ourang-outangs)”; and her discovery as their Queen, with the son taken too. AND THE ONE DETAIL THE TAILOR GOT WRONG EXPLAINS WHY FIVE SEARCHES FAILED. He wrote the colonist’s name was “either WILLIAMS, or WILLIAMSON”. The narrator is anonymous; his son is called WILLIAM. A man recalling a magazine a year later took the child’s name for the family’s. Searching for a surname that never existed is why nothing was found — and the lesson is that a negative built on a detail from memory is a negative about the memory. THE METHOD THAT WORKED, for next time: archive.org serves `https://archive.org/download/<id>/<id>_djvu.txt`. The Mirror volumes are vol 13 = `s1id13414260`, vol 14 = `s1id13414270`, vol 15 = `s1id13414280`, vol 16 = `s1id13414290`. Four volumes, 10 MB of text, searched locally in seconds. Full-text searching a downloaded corpus beats querying somebody else’s index. 21 SEPTEMBER, RE-READ AND CORRECTED — AND THE CORRECTION IS THE POINT OF WRITING METHODS DOWN. This note first recorded the volumes as `s1id134142N0` with suffixes “…260, …270”. There is no `N0` in the identifier, and every one of those four ids returns nothing. The prefix is `s1id134142` and the suffix is two digits, not three. A recipe that will not run is worse than no recipe, because the next reader spends the time before discovering it. The four ids above were fetched and the story re-read from `s1id13414270` to check every quotation in this row: the heading, the chief Karoo, “ten cows and a bull … and a choice virgin to boot”, “he foretold I should repent it”, captain Johnstone of the 72nd, the nine British soldiers, the wife Agnes — married at Beattock, which is what makes the colonist Scottish — her son William, the pongos, and her discovery as the pongos’ Queen “sent them … from the country beyond the sun”. All of it stands, and p. 318 is printed in the text itself. ONE MORE TRAP, worth more than the identifier: the Gutenberg-sourced copies of the same magazine — the `themirrorofliter…gut` items, one per weekly number — have no `_djvu.txt` at all; they hold the Gutenberg file, `<number>-8.txt`. Asking for `_djvu.txt` returns HTTP 200 and an HTML error page, which a script that only checks for a non-empty file will count as a success. Twenty-two of twenty-two “downloads” were that page. Check that what came back is the text before you search it, and treat a zero from an unverified object as no result rather than a negative. The four identifiers above were each fetched and verified on 21 September at 2.4–2.7 MB apiece, and the story is in vol 14. Noted because commit d3828d3’s message gave the same thing as a pattern, `s1id134142N0`, which reads as a literal id and resolves to nothing — the row is right and the commit message is not, and the commit is not being amended.
133. Done — finished and published this archive
Mrs Field’s daughter Mary married a Mr Bellew, a surgeon — searched, and not in the index
The first friends of the Soho household to be named anywhere. Mr & Mrs Stanhope; Mr and Mrs Warner; Mrs Field and her daughters, the eldest of whom, Miss Mary, “has married a Mr Bellew, a Surgeon, whom I do not know whether you may recollect”. A surgeon named Bellew marrying a Mary Field about 1829–30 is a findable marriage — London parish registers and the medical directories both index surgeons. This archive has had this family entirely through official paper. These are the people who came to the house. SEARCHED 20 September, AND NOT FOUND — with a proper control. Using FindMyPast’s SpouseLastName filter (a nonsense spouse returns 0, so the field genuinely restricts): FIELD × BELLEW returns ONE record in the entire Life Events collection, and it is 1854 at Stepney — twenty-four years too late, and the letter says she *had already* married by August 1830. Tried as exact spellings bellew, bellow, ballew, belew, bellewe, bellows, each against 1823–1835: all zero. Widened to 1814–1844: still zero. The name variants engine is useless here — it collapses Bellew into Bell, Boyle and Balow and returns 1,724 rows of noise. So either the marriage is not in FindMyPast’s London coverage, or the father misheard a surname he was writing from conversation. Recorded with the spellings tried so nobody repeats them.
134. Done — finished and published this archive
Frederick William Kolbe — RESOLVED: not the Settler’s son. Sibree is wrong and Schoeman is right
A flat contradiction between two printed authorities, with a chronological impossibility inside it. Sibree, entry 522: *“KOLBE, Frederick William. Born, Nov. 3, 1821, in South Africa. SON OF REV. G. A. KOLBE (No. 278). Had been connected with the Rhenish Mission… died at Cape Town, Dec. 25, 1899, aged 78.”* Author of an English–Herero Dictionary (Juta, 1883), minister at George and then Paarl. Schoeman: *“F.W. Kolbe (1821–99) van die Rynse Sendinggenootskap was nie aan hierdie familie verwant nie.”* Same man, same society, same years. AND HE CANNOT SIMPLY BE ADOPTED. Sibree’s 3 November 1821 agrees with his own age at death; the Settler’s known eldest, John Gerson Eliezar, was born 7 July 1822, and *his* age at death agrees too. Eight months apart. One of the two births is wrong, or one of the men is not who his source says. Sibree also gives him a sister — CHARLOTTE KOLBE, born 25 December 1829, who married Rev. William Sykes at Paarl in 1860; the Settler’s 1829 daughter is recorded here under another name. The paper that would settle it is at SOAS: a missionary’s own candidate papers state his parentage. RESOLVED THE SAME DAY. Three independent works: the New Catholic Encyclopedia, on his son Monsignor F.C. Kolbe — “the son of a WESTPHALIAN (GERMAN) PROTESTANT MISSIONARY who had married the daughter of a member of the London Missionary Society”; the S2A3 Biographical Database, which names him FRIEDRICH WILHELM KOLBE, Rhenish missionary and Herero scholar, married at Cape Town in 1849, at Okahandja by April 1850, fled to Otjikango when Jonker Afrikaner attacked the Hereros that August, back in the Colony from 1852 — George, Paarl, Stellenbosch, Cape Town; and DSAB Vol. 5 (HSRC, 1987), which carries its own entry on F.W. Kolbe and was in print five years before Schoeman wrote. Sibree’s register has two Kolbes in South Africa, 278 and 522, and a compiler joined them. The same species of error as Schoeman giving the father the son’s trade. Charlotte Kolbe, born 25 December 1829, goes with him.
135. Next — ready to start David
SOAS — the letter is written, with seven exact references including four journals in his own hand
Everything keeps arriving at the same door. The Council for World Missions / London Missionary Society archive at SOAS, University of London holds: (1) Incoming letters, South Africa — source 4 on Schoeman’s list, where George Augustus Kolbe’s own reports are; (2) the manuscript of the 1828 journal with Melvill, source 6, which is the same document under another name; (3) the letter Dr Philip promised in 1830 to write to him, if it was written; and (4) the candidate papers of Frederick William Kolbe, which would say whose son he was and end the contradiction between Sibree and Schoeman. Four asks, one archive, one letter — and it is in London, which is the one city on this list somebody could reach. 20 SEPTEMBER — THE CATALOGUE IS ITEM-LEVEL AND EVERYTHING IS FOUND. Sixty-two records on the surname. Four journals: CWM/LMS/05/05/100 (Melvill and Kolbe, tour north-east of Philippolis, 13–22 Nov 1828 — the manuscript behind source 6), /102 (1828–1831), /106 (1831–1834), /108 (6 Oct 1832 – Sep 1833). Incoming correspondence: /044 (1835, and the catalogue names *“George Augustus Kolbe, Philippolis”* among the correspondents), /046 (1836), and the 1837 file — 1836–1837 is Box 15, which is the accusation and the dismissal. Drafted as `requests/soas-cwm-kolbe-journals.md`; the sending is David’s. `special.collections@soas.ac.uk`, 10 Thornhaugh Street, Russell Square.
136. Next — ready to start Western Cape Archives
Three Cape Archives CO references for the Nautilus — and one of them is a signed petition
Nash cites three sources for Owen’s party, all at the Western Cape Archives, which this archive is already writing to: CO 6137/1,4 — the list of settlers under Thomas Owen’s direction; CO 8543 — Special Commissioner Hayward’s notes, which is where “this querulous man” comes from; and CO 3917 — the **Memorial of independent settlers aboard the *Nautilus*. A memorial is a petition and a petition is signed. If George Augustus Kolbe signed it he was seventeen, and it would be the earliest example of his own hand in existence. Nash also records that no Agent of Transports’ Return has ever been traced for the *Nautilus*, so these three are the whole documentary basis for the roll. Add to the standing Western Cape Archives chase.**
137. Next — ready to start this archive
Dictionary of South African Biography Vol. 5 — and the series was terminated in 1987, five years before Schoeman wrote
Found in S2A3’s source list while resolving the F.W. Kolbe question. Two reasons to want it. First, it would give the Rhenish missionary’s parentage and birthplace outright, closing the Sibree error with a named source instead of an inference from three. Second, and larger: DSAB Vol. 5 was published in 1987 and Schoeman’s G.A. Kolbe draft is dated February 1992, so the Settler’s entry was written for a later volume — and the copy Melanie Geustyn sent is headed *voorlopige teks*. If the later volumes never appeared, that typescript is the only form the entry exists in, which is exactly the question the thank-you letter now asks her.
21 SEPTEMBER — THE VOLUME IS LOCATED AND IT IS DIGITISED. *Dictionary of South African Biography* Volume 5 is on the Internet Archive as `dictionaryofsout0005unse` — 978 pages, publisher *Nasional Boekhandel Bpk. for the National Council for Social Research*, which matches. It is not open full text: its collections are `inlibrary` and `printdisabled`, so it is a borrowable scan. The OCR exists — the text file is 6.09 MB — and is searchable from inside the reader once borrowed. So this no longer needs a library building. A free Internet Archive account borrows it an hour at a time, and the F.W. Kolbe entry can be read and searched there. A warning about my own first attempt, because it is the fault this archive keeps finding: fetching the `/stream/.../_djvu.txt` URL returned an HTML wrapper, I counted zero occurrences of *Kolbe* in it, and that zero meant nothing whatever — the `<pre>` was empty because the text loads by script. A null from the wrong object is not a null. Nothing about the entry's contents has been read and nothing is claimed about it. 21 SEPTEMBER, LATER — THE SECOND AND LARGER QUESTION IS ANSWERED, AND NOT BY BORROWING ANYTHING. The successor volume says it outright. **E. J. Verwey (ed.), *New Dictionary of South African Biography*, Volume 1 (HSRC Publishers, Pretoria, 1995) — foreword by Nelson Mandela, dated September 1994 — is openly readable on the Internet Archive as `bub_gb_rl8nkyID3WsC`** (`_djvu.txt`, 1.06 MB, collection `americana`). Its Introduction: *“The decision to terminate the DSAB after the publication of Volume V in 1987 (at that stage 4 518 biographies had been published) and to embark on the New dictionary of South African biography (NDSAB) was taken for two reasons — firstly, to adapt to changes in the external environment that seriously reduced the resources available for the continuation of the programme (in fact, threatened its very survival). Secondly, there was a need to reconsider the programme and correct some imbalances.”* THERE IS NO DSAB VOLUME VI AND THERE NEVER WAS. Schoeman's G.A. Kolbe draft is dated February 1992 — five years after the series it was written for had been stopped, and three years before the replacement appeared. The replacement is not a continuation in the same numbering: NDSAB I carries 129 biographies against the DSAB's average of 900 a volume, and the Settler is not among them. **So the *voorlopige teks* Melanie Geustyn sent is, on the published record, the only form that entry has ever taken — which is what the letter asks her, and the archive can now say why it is asking. AND THE SAME BOOK PRINTS A CUMULATIVE INDEX TO DSAB VOLUMES I–V, WHICH CORRECTS THIS ROW. This row assumed one Kolbe entry. There are three: KOLBE, FRIEDRICH CARL — Vol. I, p. 433; KOLBE, FRIEDRICH WILHELM — Vol. V, p. 419; and KOLBE, FREDERICK FORTUNATUS HOFMEYR — Vol. V, p. 419. (Separately indexed, and not one of ours: Kolb(e)(n), Peter — Vol. III, p. 475, the eighteenth-century Cape writer.) TWO OF THE THREE ARE ON THE SAME PAGE, which is how the DSAB sets a family down together — so page 419 of Volume V is now the single page most worth reading in the whole series, and one borrowing gets both. AND THEN THE FIRST QUESTION ANSWERED ITSELF TOO, WHICH IS THE BETTER RESULT AND THE LESS COMFORTABLE ONE.** This row wanted Volume V because it *“would give the Rhenish missionary's parentage and birthplace outright, closing the Sibree error with a named source”*. The birthplace is now named without opening the volume, and it rules him out of this family. FRIEDRICH WILHELM KOLBE, 1821–1899, was born at GÜTERSLOH in WESTPHALIA and died at CAPE TOWN (Wikidata Q25615009, resolving both places). He was of the RHENISH Missionary Society, married ISABELLA MARIA ELLIOTT, daughter of the Rev. William Elliott, at Cape Town in 1849, went to Okahandja in 1850, was driven out by Jonker Afrikaner's attack on the Hereros that August, and came back to the Colony in 1852 — George, Paarl, Stellenbosch, Cape Town (S2A3, entry for Mrs Isabella Maria Kolbe, compiled by C. Plug, which cites DSAB Vol. 5 for him). Their son is Monsignor F. C. Kolbe, 1854–1936, the theologian and writer. A GERMAN-BORN WESTPHALIAN OF THE RHENISH SOCIETY IS NOT THE SETTLER'S KIN. The Settler was born in Soho in 1802, son of a Württemberg tailor settled in London, and he was LMS. Nineteen years apart, two countries of birth, two societies. They share a surname and a mission field and nothing else. A LEAD I RAISED AND KILLED WITHIN THE HOUR, WRITTEN DOWN BECAUSE THAT IS THE POINT. The cumulative index reads, in OCR, *“Kolbe, Frederick Fortunatus Hofmeyr, V, 419”* beside *“Kolbe, Friedrich Wilhelm, V, 419”* — and FORTUNATUS is a name this archive traces across three families. The Dictionary of African Christian Biography then says Monsignor Kolbe's parents *“were members of the Protestant London Missionary Society”*, and the Settler was LMS, so for about twenty minutes there was a Fortunatus and an LMS household pointing the same way. Both halves fail, and my first reading of the index was wrong too. I took the two entries at p. 419 for father and son. That will not hold, because the Monsignor is F. C. — Frederick Charles — and the index has a KOLBE, FRIEDRICH CARL at Vol. I, p. 433, and Volume I (1968) covered people who died before 1950, which he did, in 1936. The son is very probably the Vol. I entry, not the second Vol. V one, which leaves the p. 419 pair unexplained and a *Fortunatus* in the index still unaccounted for. What does not depend on any of that: the father is Westphalian and Rhenish; the DACB's LMS is loose — it is the mother's father, the Rev. William Elliott, who was LMS — and the DACB also dates the Monsignor 1864 where Wikidata and S2A3 give 1854. The Monsignor's own initials are F. C., not F. F. H., everywhere he is named by people who knew the family, so the index line is probably two entries run together by the scanner or a name this archive has misread. A forename and a society are not a descent, and an OCR'd index is not a document. WHAT IS LEFT, AND IT IS SMALLER THAN THIS ROW BEGAN. Volume V, p. 419 would still confirm F.W. Kolbe's parentage from the printed page and settle what that third index line actually says; Volume I, p. 433 holds a KOLBE, FRIEDRICH CARL nobody here has accounted for. Both are borrowable, not open, and borrowing needs a signed-in Internet Archive account, which is not this archive's to create. That remainder stops at David. What no longer needs anybody is the question this row was really built on: the DSAB stopped at five volumes and the Settler's entry was never printed.
138. Blocked — waiting on a person or an institution this archive
Read the four Kolbe journals when they come, and test Schoeman against them
This is the row everything today was pointing at. Four manuscripts, 1828–1834, in George Augustus Kolbe’s own hand. The sum of his own words in this archive is currently one sentence, quoted at second hand through Legassick. When they arrive they will test, from the inside, almost everything this archive has published about him on one source: the Graaff-Reinet years, Boesmanskool and James Clark, the Caledon journey, the move to Philippolis, Melvill’s departure, the ordination, the Berlin Missionary Society at Bethanie — and they may say something about 1836. They will also be read for what they say about the people among whom he worked, who appear in this archive so far only as the objects of other men’s sentences.
139. Done — finished and published this archive
Re-justify all six Pretoria estate files against what the archive now holds, before any of them is paid for
The 15 September brief asked for six death notices “for the parents, the birthplace, the occupation and the list of children”. Since then the Dutch Reformed registers have delivered the parents of that whole generation, so for several files the stated reason no longer stands. A quick audit: Henry James Booyzen (MHG 10/57) has his baptism, marriage, death, burial and a corrected cause — *low*. Catharina Maria Sophia Daschner (MHG 3734/46) has her baptism, her husband’s read estate and her grave; what is missing is her exact death date and her children in 1946 — *moderate*. George Downing Mountjoy (MHG 1095/48) has dates, burial, occupation and marriage; missing is his children — *moderate, and he is in the direct line*. John Barry (MHG 769/52) has a gravestone giving 26.4.1884–23.1.1952 and “aan ons vader”, and not one child named — *good*. Gert Jacobus (MHG 5155/51), from the second marriage, is the thinnest — *good*. MHG 52653 is a doubtful reference and expected to fail. Nothing should be paid for on a reason that has already been answered elsewhere. DONE 20 September, and it removed two files from the order. Each was tested against the free trees, at David’s prompting. MHG 769/52 (John Barry) — ANSWERED FREE: FamilySearch M87R-4X4 gives birth 26 April 1884 and death 23 January 1952 at DELMAS, matching the Brakpan gravestone to the day, plus a wife Margritha Maria Susanna Poole (1886–1947) and TEN CHILDREN — the two eldest named Willem Hermanus and Catharina Maria Sophia, his own parents’ names, which is this family’s documented naming pattern. MHG 3734/46 (Daschner) — ANSWERED FREE by her 1891 Barkly Oost baptism, held here. MHG 5155/51 (Gert Jacobus) — NOT ANSWERED: FamilySearch G5DK-WZS gives parents only, no death, no spouse, no children. MHG 1095/48 (George Downing Mountjoy) — NOT ANSWERED and now sharper: the MyHeritage capture gives his death as 21 February 1947 where this archive has 11 December 1947 from the burial register, with a burial on 14 December. Only the death notice settles which. MHG 40293 (Kolbe 1918) — UNKNOWN: not found, but FamilySearch tree search is scored and not filtered (27,656 and 110,097 results on narrow queries), so absence there proves nothing. A tree gives names to look for; a death notice gives proof. Two files were buying names this archive can already read for nothing.
140. Next — ready to start this archive
The Geni tree tested — and its last unproved name, General Willem Johannes Kolbe, turns out to be published and already here
Twelve names for a household this archive has been unable to name since the Commandant’s obituary was read. The rule is the one already written for Schoeman’s bibliography: a list worth working is not a witness. Nobody is entered here on a tree’s word. Start with the two that are already half-proved — Rosa Magdalene Kolbe (1890 Bloemfontein baptism, held) and Maria Margaretha Downing Rousseau (TAB MHG 5450/68, held) — and then General Willem Johannes Kolbe, because a general leaves a paper trail a farmer does not. Beware the name collisions: this Frederik Fortunatus (b. 1857 Bloemfontein, d. 1930) is a fourth man of the name, distinct from the Settler’s tenth child (1839 Bethulie – 1929 Utrecht) and from Frederick Fortunatus Downing Kolbe (d. 1918, TAB MHG 40293). This archive withdrew a claim within the hour once before over two men sharing a name in one town. 21 SEPTEMBER — THE TREE IS HOLDING UP WHERE REGISTERS CAN TEST IT. Two of the named siblings are now in civil and church registers with McCabe wives: JAN HENDRIK KOLBE (m. Rose Maria McCabe, Bloemfontein, 4 June 1872, b. 1849) and FREDERIK FORTUNATUS (m. Maria Eliz Petronella Antoinette McCabe). Jan Hendrik’s eldest son is JOHN GERSON ELIEZER, the Commandant’s name entire — which is register evidence of the descent the tree asserts. Still not adopted; still awaiting VAB MHG M777. Next from the tree: General Willem Johannes Kolbe, for whom nothing published can be found — ‘General’ may be an honorific. 21 SEPTEMBER, LATER — THE LAST NAME ON THE TREE IS THE ONE THAT BREAKS IT OPEN, AND HE WAS ALREADY IN THIS ARCHIVE. This row asked whether General Willem Johannes Kolbe was real or whether ‘General’ was an honorific, because nothing published could be found for him. He is real, he is published, and the rank is his. *Afrikanergeskiedenis* (Petronel Fourie) and the Afrikaans Wikipedia both carry him, and behind them stands a printed reference work: **P. J. Nienaber, *Afrikaanse Biografiese Woordeboek* (Voortrekkerpers, 1947), p. 139. Born 9 July 1847 in the district of PHILIPPOLIS; led the Bloemfontein commando; commanded the southern districts of Bloemfontein; made general during the siege of Kimberley; at Paardeberg, Driefontein and the Brandwaterkom; and in De Wet’s invasion of the Cape Colony in February 1901. Two horses were shot under him. The French general de Villebois-Mareuil praised his leadership; he was liked in his commando. He died in 1941, aged ninety-four. AND THE SAME ARTICLE NAMES HIS HOUSE, WHICH IS WHY THIS IS NOT A NAME MATCH. Where the pedigree comes from, stated exactly: it is set out on the Afrikaans Wikipedia article, which lists two sources and footnotes neither claim to either. Of the two, only Nienaber can be carrying it** — the *Afrikanergeskiedenis* piece was read here in full and contains no parents, no grandparents and no farm, only the military career. So the genealogy is Nienaber 1947 by elimination, and has not been read from the page of Nienaber itself. That matters, and it is why nothing below is adopted. It gives his father as J. G. E. Kolbe, his mother as ANNA JACOBA VAN DER MERWE, his grandfather as GEORGE AUGUSTUS KOLBE, a German missionary, his grandmother as MARGARET DOWNING, the grandfather’s place as Knapdaar, Burgersdorp, and the home farm as ONDER-LEEUWBERG in the district of Bloemfontein — and it says his father was a commandant in the Basuto war of 1865. Every one of those is already held here, independently. The Commandant’s obituary gives the marriage to Anna Jacoba van der Merwe; the Settler’s obituary gives him as a Londoner of a Württemberg house who named his farm *Wurtemberg*, which is exactly how a 1947 Afrikaans lexicographer would come to call him German; the Burgersdorp stone gives Margaret Downing and Burgersdorp; and Leeuwberg is the farm the Commandant died on. Father, mother, both grandparents, the grandparents' town, the home farm and the father's rank — every one of them already held here, from obituaries and a gravestone, before anybody had heard of this entry. Two of the agreements are the forenames of women and one is the farm the Commandant died on. That is a household, not a coincidence of surname. THE MAN WAS ALREADY IN THIS ARCHIVE AS A NUMBER. *Williem Johannes Kolbe of Panfontein*, taken prisoner in 1901 aged 54 in the Bloemfontein district, TNA WO 108/368 folio 603 — the right name, the right year of birth, the right district, and captured in the year of the Cape invasion he is published as riding on. This archive held the General for weeks and did not know it, because a prisoner-of-war list gives a name, an age and a farm and nothing to hang them on. Still not entered as the Commandant’s son on Nienaber’s word alone — but this is a printed biographical dictionary naming four ancestors, not a user tree, and it is a different class of thing from what this row was warning against. VAB MHG M777 remains the paper that would settle it, and there is now a second — see the row below on his own memoir. 21 SEPTEMBER — NIENABER IS NOT DIGITISED, AND THAT IS A CONTROLLED NEGATIVE. **P. J. Nienaber, *Afrikaanse Biografiese Woordeboek* (Voortrekkerpers, 1947) returns 0 on the Internet Archive, on a title search and on a loose search. Three controls before believing it: the DSAB title search returns 5 including `dictionaryofsout0005unse`, the identifier row 137 reached independently; `title:("Afrikaanse")` returns 570, so Afrikaans is not invisible to the index; and Voortrekkerpers returns 2 items, so the publisher is not absent. So p. 139 cannot be read the way DSAB Volume V can. DSAB V is borrowable with a free account; this is not there at all, and the realistic holders are the NLSA by legal deposit and South African university libraries — the same shape of problem as row 117's Olivier book, and possibly the same letter. NOTHING ABOUT THE PEDIGREE CHANGES. It is neither confirmed nor withdrawn. It must not be upgraded to documented until somebody reads that page, and this row is now clearer about how far away that page is. 27 SEPTEMBER — THE TWELVE ARE COMING OUT OF REGISTERS, NOT OUT OF THE TREE. Searching the Dutch Reformed baptisms on ANNA JACOBA VAN DER MERWE — her husband's name is spelled six different ways across six documents and cannot be searched — returns seven baptisms and two civil death registrations naming both parents. JAN HENDRIK IS NOW DOCUMENTED (bapt. Fauresmith 1849), WILLEM JOHANNES IS NOW DOCUMENTED (bapt. Fauresmith 1848), and ANDRIES FREDERICK OF LEEUWBERG is his father's son by his own 1918 death registration. Four are entered; the rest are listed at /searched/ and deliberately not yet made into people. VAB MHG M777 is still wanted — it would give all twelve at once, with the four daughters — but it is no longer the only road. 27 SEPTEMBER, SECOND PASS — THE GENERAL IS NOW IN A CIVIL REGISTER. His death registration gives 23 April 1941 at Meriba, Bloemfontein, born 1848 at PHILIPPOLIS** — the dictionary's death year and the dictionary's birthplace, reached by a registrar who never read it. With the Fauresmith baptism of 1848 that is three documents converging on one man.
141. Done — finished and published this archive
The Kolbe–McCabe “coincidence” — DONE: it is two marriages, not one, and both are in the registers
This archive wrote of the 1890 Bloemfontein baptism: *“a second Kolbe–McCabe marriage in the same small Free State world, and this archive does not yet know whether that is a coincidence.”* The Geni profile suggests it is not. The Commandant’s widow Anna Jacoba van der Merwe married Francis McCabe, Landdrost of Bloemfontein, about April 1871; and her son Frederik Fortunatus married Maria Petronella Elizabeth Antoinette McCabe. Stepmother and daughter-in-law out of the same family. McCabe died in February 1872 leaving “four sons and two daughters, besides a large family of step-children” — so the two households were one for eleven months, and a marriage between them afterwards is exactly what one would expect. WHAT WOULD PROVE IT: the marriage record of Frederik Fortunatus Kolbe and Maria Petronella Elizabeth Antoinetta McCabe, and Francis McCabe’s own estate naming his children. ANSWERED 21 September, from civil and church registers rather than trees. South Africa, Civil Marriage Records, 1801–1974: *ANNA JACOBA KOLBE, bride, b. 1825, spouse FRANCIS McCABE, Free State, 13 JULY 1871*; and *JAN HENDRIK KOLBE, groom, b. 1849, spouse ROSE MARIA McCABE, Bloemfontein, 4 JUNE 1872*. Free State Dutch Reformed Church Records carry both couples as parents — Frederik Fortunatus with Maria Eliz Petronella Antoinette McCabe, and John Henry (= Jan Hendrik) with Rosa Maria McCabe, whose eldest son is named JOHN GERSON ELIEZER, the Commandant’s name entire. TWO OF THE COMMANDANT’S SONS MARRIED TWO OF FRANCIS McCABE’S DAUGHTERS. Two is not a coincidence; it is a stepfamily marrying within itself, out of a household that lasted seven months. And Jan Hendrik married four months after his stepfather died. A DATE THIS ARCHIVE HAD ONLY BY INFERENCE IS NOW WRITTEN DOWN: the obituary’s “about ten months since” gave April 1871; the register says 13 July 1871, which is seven. These are index entries, not images read — the images would give ages, fathers and witnesses.
142. Done — finished and published this archive
Margaret Downing’s mother — DONE: both candidate surnames tested and neither survives
The FamilySearch Family Tree gives Margaret Georgina Elizabeth Downing’s parents as JOHN PAUL DOWNING (1765–1839) and MARY KING POOLEY (1768–1834). This archive has had a father’s name from the tree for a long time and never a mother at all — her 1867 Burgersdorp death notice answers the parents question with “Both dead” and no names. FROM A USER TREE AND NOT ADOPTED. But “King” carried as a middle name is how a family carries a mother’s maiden name, exactly as with Mary Ann Maria *Ford* Kolbe, so the name is structured like a real one. A Downing–Pooley marriage in London in the 1790s would be a document, and so would a baptism for Margaret in 1803 naming both parents. Look for both. 20 SEPTEMBER, SEARCHED — AND THE PREMISE WOBBLED. FindMyPast’s Life Events collection holds exactly one Downing–Pooley marriage before civil registration: JOHN DOWNING m. MARY POOLEY, 10 MAY 1802, WELCOMBE (Bishop’s Transcripts), DEVON — nine months and thirteen days before Margaret’s birth, and in the county this archive’s own tree puts the Downings in (Trick, Heard, Hooper). But Margaret was born in LONDON, no 1803 London baptism for her can be found, and the two trees disagree about the woman’s surname — FamilySearch says Mary King Pooley, this archive’s own tree says Mary P King. The “P” may be Pooley, or Pooley may be a reading of the “P”. A candidate, not a finding; nothing adopted. CLOSED 21 September. The single Downing–Pooley marriage in England is Welcombe, Devon, 10 May 1802, and that couple stayed in Devon — a child baptised there in 1803 and more in 1806, 1809 and 1810, where Margaret was born in London. Ruled out. No London baptism exists for her, and that is a real negative because the index covers London well (37 of 100 Smith baptisms in the same window are London). And the tree’s “Pooley (1768–1834)” is very probably a reconstruction built from that Devon marriage plus a Mary Downing buried at Welcombe in 1834. She has no mother, and that is now a tested position.
143. Done — finished and published this archive
Two trees give the Commandant’s son two birth years — 1857 and 1859
Geni gives Frederik Fortunatus Kolbe born 22 October 1857 at Bloemfontein; the FamilySearch Family Tree gives 1859. Both agree he died in 1930. FROM USER TREES, NOT ADOPTED, and the disagreement is the useful part — it shows the two are not independent witnesses. Where collaborative trees agree it may be because one copied the other; where they differ, neither settles anything. The Bloemfontein baptism register would. His own children are better documented than he is: Rosa Magdalena, baptised Bloemfontein 1890, is already held here. 21 SEPTEMBER, ATTEMPTED AND BLOCKED BY THE THING IT IS ABOUT. A record search for *Frederik Fortunatus Kolbe* born 1855–1862 in South Africa returns the wrong man every time — the Settler’s tenth child, born 1839 and married to Elizabeth Johanna du Toit, whose eight children this archive already holds. The name collision is itself the obstacle, and the date filter does not restrict enough to cut through it. The Bloemfontein baptism register browsed by year is the way, not a name search. 27 SEPTEMBER — ANSWERED, AND THE ANSWER IS THAT THE DOCUMENTS DISAGREE. The gravestone gives 22 October 1859; the civil death registration gives age 73 at death on 18 December 1930, which is 1857. Each tree was following one of them, so the row's own reasoning — that the disagreement proved the trees were not independent — is withdrawn at /corrections/. The baptism is not in the index and the stone's photograph is paywalled. What the search did produce is seven of the Commandant's children from fonts, by searching on the mother's name.
144. Done — finished and published this archive
Test the Welcombe marriage of 10 May 1802 — DONE, and it is ruled out
Three ways to settle it, in order of decisiveness. (1) A London baptism for Margaret Georgina Elizabeth Downing in 1803 naming both parents — not found on FindMyPast, so try the London Metropolitan Archives registers and the Westminster sets this archive already used for the Kolbe baptisms. (2) The Welcombe register itself rather than the bishop’s transcript — it may give ages, conditions, fathers or witnesses, and the Devon Heritage Centre holds it. (3) A burial for Mary Downing about 1834, the year both trees give her. If the Welcombe couple stayed in Devon, they are not this couple, and ruling it out is as useful as confirming it. 20 SEPTEMBER, ROUTE (1) TRIED AND IT FAILED FOR A REASON WORTH KNOWING. No London baptism for Margaret Downing exists in FindMyPast for 1801–1805 — and after a proper control the nil is real: of 100 SMITH baptisms in the same window thirty-seven are London, Middlesex or Westminster, so the coverage is excellent and the absence means something. Either she was not baptised in an Anglican London parish, or her family was not a London family — and the second reading strengthens Welcombe, because a couple married in North Devon in May 1802 with a child born in London nine months later and no London kin in the registers is exactly what newly arrived country people look like. Also tested and failed: the idea that the surname’s own geography supports Devon. A 160-record sample runs Yorkshire 20, Suffolk 16, Norfolk 16, Cornwall 14, Devon 5 — Downing is a national name, and the Devon argument rests only on the tree’s specific ancestry. RULED OUT THE SAME DAY, by asking whether the couple stayed. Every Downing record in the parish: 1802 John marriage, 1803 John BAPTISM, 1806 William, 1809 Emmey, 1810 Elizabeth — they baptised a child at Welcombe the year after the wedding and kept doing it, and the Pooleys are equally rooted there. Margaret was born in London. And the 1834 Mary Downing burial in that same parish is almost certainly where the tree’s “Pooley” and its “1834” both came from — a previous researcher running this identical search and stopping one question earlier.
145. Done — finished and published this archive
King or Pooley — DONE: neither, on present evidence
Two user trees, one woman, two surnames. The FamilySearch Family Tree: Mary King Pooley (1768–1834). The tree this archive was built from: Mary P King (1768–1834). Same years, different surname — which means at least one of them has parsed a name wrongly, and it is probably an initial being expanded or a surname being demoted. FROM USER TREES AND NOT ADOPTED. The whole Pooley search rests on this being right, so it should be settled before more time goes into it. The document is Margaret’s 1803 baptism, or her parents’ marriage under either name. Note that a Downing–KING marriage has not yet been searched at all, and that is the obvious next sweep. 20 SEPTEMBER, ANSWERED IN SUBSTANCE IF NOT IN NAME. The Welcombe test suggests neither surname is hers and that “Pooley (1768–1834)” is a reconstruction built from one Devon marriage and one Devon burial. So the question is no longer King or Pooley. It is whether either name has any record behind it at all — and on present evidence neither does. Margaret Downing has no mother, and that is now a documented position rather than a gap.
Closed 21 September. The row asked which surname was hers and the answer is neither, on present evidence — that is a decision, not a deferral, and the row was still reading *next* while its own note said *ANSWERED IN SUBSTANCE*. Checked before closing: `/documents/#downing-pooley` states the whole of it — the date works to nine months and thirteen days, the county agrees with this archive's own Devon tree without either knowing about the other, no London baptism exists for her in 1803, and the two trees cannot agree whether she was a King or a Pooley. `people.json` holds neither name, so nothing was adopted. Margaret Downing has no documented mother, and that is now a published position rather than an open question. What would reopen it is a document: her 1803 baptism naming both parents, or a Downing–KING marriage, which has still never been swept. CLOSED 21 September. Neither surname survives testing. Pooley traces to a Devon couple who never left Devon. King returns, on the SpouseLastName filter for 1790–1806, one Norfolk event indexed three times — a *Mary* Downing at King’s Lynn in 1790, wrong person and wrong direction. The question is no longer which of the two; it is that neither has a record behind it.
146. Done — finished and published this archive
Build ONE settled Pretoria list — every item checked against free sources before Anne is asked to price it
This is now owed to somebody. The letter of 20 September promised Anne Brest *“one list, already checked against what is freely available, so that everything on it is something only the archive can answer”*. That promise is the whole reason her Tuesday was cancelled, and it has to be kept properly or not made again. The method, which should have been applied from the start: for every reference, state what is actually missing; search FamilySearch’s tree, Geni and the MyHeritage capture for it; and keep the item only if those come back empty or only if a document is needed to prove what a tree merely asserts — and say which of the two it is. The three that survive so far are TAB MHG 40293 (Kolbe, 1918), MHG 5155/51 (Gert Jacobus) and MHG 1095/48 (George Downing Mountjoy, whose death date the tree and the burial register disagree about). The nineteen items of Groups A, B and C have never been through this at all — seven WLD court files, five Lerena items, the OD signatures, the CJC Marico claim, SAB VWR 2. None of them goes back on a list until it has been. DONE 20 September — and it lives on /errands/, not in a note file. The Pretoria entry there was already the right structure and was stale in exactly the predicted way: it still called the Daschner notice *“the most wanted page in this list”* for parents this archive had held for a fortnight. Rebuilt in three tiers. (1) The three death notices that survived the check, each labelled with why — MHG 40293 (untested, because FamilySearch tree search is scored not filtered), MHG 5155/51 (the tree has parents only, nothing else), MHG 1095/48 (kept for a NEW reason: the tree says he died 21 February 1947 and the burial register says 11 December, and only the notice settles it). (2) The items no tree could ever answer — the WLD divorce, the seven CJC Marico claims, the two OD school files wanted for a signature, SAB VWR 2, and the David Schalk index search. These were never candidates for the test. (3) Dropped, with reasons, so nobody orders them again — MHG 3734/46, MHG 769/52, MHG 10/57 and MHG 2703/63. MHG 52653 was reclassified from a file to an index question: ten minutes at a counter for the correct reference, not an hour in a strongroom. Groups B and C were not audited and are not on it — Wear, Longhurst and Rogue Luis Lerena return zero hits in this archive's people, so they belong to the Wear research and the Lerena archive and need their own owner. 22 SEPTEMBER — 52653 IS A FILE AND THIS WAS WRONG. Anne Brest had located it on NAAIRS on 15 September and said so, with a link: the absent year suffix is normal for some 1920s references. She was told twice not to hunt it after she had already found it. The correction.
147. Next — ready to start David
Groups B and C need an owner — twelve Pretoria items that are not this archive’s to audit
They were on the order to Anne Brest and they were never checked, because this archive cannot check them. Wear, Longhurst and Rogue Luis Lerena return zero hits in `people.js`; Nuno Fernando Lerena has one line. Group B is seven items — WLD 2026/1951, 2960/1962, 185/1949, 790/1957, 978/1960, 2225/1953 and MHG 3632/50 — and belongs to the Wear/Luwinski research. Group C is five — TPD 1367/1980, 1248/1979, MHG 1518/73, 6746/71, WLD 8043/77 — and belongs to the Lerena archive, which is a sibling site with the same method. Most of them are court files and wills, which no family tree holds, so they would very likely survive the same test that cut this archive's list in half — but somebody who holds those families has to run it, and it should be run before they go back on any invoice. The promise made to Anne was one checked list; a list half-checked is not that.
148. Done — finished and published this archive
All seven errand entries audited for staleness — four were out of date
/errands/ is the page somebody standing at a door reads, so a stale entry there wastes a journey rather than a search. All seven audited on 20 September. PRETORIA — rebuilt entirely; it still called the Daschner notice “the most wanted page in this list” for parents held here a fortnight. WESTERN CAPE — still asked somebody to stand at a door for MSB 751, answered by the National Library on 18 September; marked answered with the result recorded rather than deleted, because an errand that quietly vanishes teaches nothing. KEW — said “not yet written to” when the letter has been written and corrected twice, and asked for “WO 12” when the piece is now WO 12/1450 and the window four quarters wide, because the regiment left the Cape in September 1817 and not 1816. BARKLY EAST — the Indwe index has since been read and the entry did not say so; sharpened to record what it gave and what only the originals will. GRAHAMSTOWN — gained a second institution: the Albany Museum, which has held a photocopy of the 1819 Kolbe–Downing marriage certificate since before 1992. PIETERMARITZBURG and BLOEMFONTEIN were current; Bloemfontein’s C129 was sharpened instead, because it is now the document that would prove a stepbrother married a stepsister.
149. Next — ready to start this archive
Württemberg before 1792 — the tailor’s marriage is probably not in England
The negative is now controlled and it points abroad. No Kolbe–Larman marriage exists in FindMyPast in either direction, with a working control; and the index demonstrably reaches this family’s own parish, because it holds the 1810 Kohlbeck banns at St Anne’s Soho. The German Lutheran register is ruled out by statute — Hardwicke’s Act of 1753 required an Anglican ceremony for validity, with exemptions only for Quakers and Jews. So: an unindexed English parish, a wrong surname, or a marriage abroad. He was born in Württemberg in 1765, entered his birth county as “FOREIGN” in the 1841 census, and the first child of the house was baptised in London in 1792, when he was twenty-seven. The gap between birth and first baptism is twenty-seven years and this archive has nothing in it. Where to look: the Württemberg church registers, which are well digitised — FamilySearch and Archion both hold them — and the family tree’s own claim of a line back to a Johann Gottlieb Kolbe of 1699 and Maria Barbara Jung, which is tree-only and untested and would give a parish to start in. A marriage abroad would also explain why no English record has ever been found for a man this archive can otherwise follow through 121 rate books. 21 SEPTEMBER, FIRST PASS WORKED. FindMyPast is the wrong tool and that is worth writing down: it indexes *Württemberg* almost entirely as a birthplace field in emigration records — American naturalisation petitions, passenger lists, English censuses — not as German parish registers. `Kolbe` + Württemberg returns 0 there against a working control. FamilySearch is the right tool and it half-answers. Its *Germany, Lutheran Baptisms, Marriages and Burials, 1500–1971* and *Germany, Deaths and Burials, 1582–1958* do reach Württemberg, and Kolbes are plainly there — Albershausen, Altdorf (Nürtingen), Knittlingen (Maulbronn), Sulz, Eltingen (Leonberg), across the seventeenth and eighteenth centuries. So the family tree’s “Württemberg” is at least a plausible region, which it had never been shown to be. But there is no Gottlieb or Gottlob Kolbe indexed in Württemberg at all — three results, none relevant. THE CAVEAT THAT MATTERS: this is a search of an INDEX, not of the registers. Much of the Württemberg church-book material on FamilySearch is browse-only images, and Archion holds the rest behind a subscription. NEXT: the named parishes above, browsed, and Archion. Noted in passing: a Carl Gottfried Kolbe buried at Sulz, Württemberg, 10 September 1787, born 1736 — the tree gives the Settler’s grandfather as born 1736. Wrong forename, wrong death year, right decade and region. Not a match and not adopted. 21 SEPTEMBER, SECOND PASS — THE FINDMYPAST NEGATIVE IS WITHDRAWN AND REPLACED WITH A NARROWER ONE. The claim above that FindMyPast indexes Württemberg only as an emigration birthplace is wrong, and the correction is logged. FindMyPast holds *Germany Marriage Index 1558–1929*, a real German marriage index with 338 Kolbe marriages, giving both parties, the year and the parish — and it reaches Württemberg: Grötzingen, Schwarzwaldkreis 1685 and Knittlingen, Neckar 1741, which is one of the very parishes the first pass named. The 0 came from a place filter that returns 0 to everything, including `westfalen` when seven of eleven records in the window are Westphalian. WHAT IS ACTUALLY TRUE NOW: all eleven Kolbe marriages in the index between 1778 and 1798 were read, and none is a Gottlieb and none is in Württemberg — the window's coverage is entirely Prussian, Pomerania to the Rhineland. So the index cannot answer this question in this period, rather than answering it in the negative. NEXT, AND UNCHANGED: the named parishes browsed on FamilySearch's images, and Archion. But the search is one step less blind: Knittlingen is now a parish where Kolbes demonstrably married, and it is on the first-pass list for independent reasons. AND THERE IS NOW A NAMED COLLECTION TO READ, WHICH THERE WAS NOT THIS MORNING. *Germany Birth and Baptism Index 1558–1898* on FindMyPast holds 952 Kolbe entries, and a baptism entry gives the parish and the parents — which is exactly what is missing for the twenty-seven years between his birth in 1765 and the first London baptism in 1792. The 952 have NOT been read and nothing is claimed about them; FindMyPast stopped rendering after a run of rapid queries and the reading is the next session's. Two traps to carry into it, both measured: the place filter on these German sets returns 0 to everything, including places printed in its own results, so the 952 must be sorted by eye and not filtered; and `datasetname` and the year filters annihilate each other — 952 alone, 0 with a year added, while `sourcecategory=parish baptisms` with the same year works and returns 619. Filter by category and year, then read the record set off each row.
150. Done — finished and published David
Pay Invoice 464 — R5,700 plus R200, halved and credited against the next list
SHE REPLIED IN FIVE HOURS AND SHE WAS RIGHT TWICE. The “several working days” transfer argument was this archive’s and it does not survive contact with somebody who takes foreign payments every week: *“BANK payments from Australia usually takes two days and I sent you the Invoice a day or two ago.”* Invoice 463 is dated 16 September. Paid when promised, it would have arrived. And on notice: her message came at 21:21 Saturday and the stand-down at 02:03 Sunday. She has spent real money — an assistant paid upfront for Tuesday — and real standing, having arranged as a favour for the files to be pulled in advance at preservation. She proposes a part invoice credited against the new list, which is fair and is less than she could ask. The reply is drafted and accepts it without argument. When the figure arrives it should be paid promptly and without a second promise of speed that is not kept, because that is the thing that actually went wrong here. 20 SEPTEMBER, ANSWERED AND SETTLED. INVOICE 464: R11,400 ‘Reduced to half’ = R5,700, plus R200 for the South African deduction — R5,900, about A$510 — and she will deduct it from the new list, so it is a credit and not a loss. The free file still stands. She named her assistant: Tebogo, English name Colman. She closed with *“thanks for being such a gentleman about this”* and *“I am looking forward to continuing on with you”*. Pay it promptly and without being asked twice — that is the whole repair. She is back in the office on Tuesday. PAID 21 SEPTEMBER 2026 — OFX deal 9577565A. R5,700 plus the R200 South African transaction deduction, R5,900, about A$510, and by her own proposal it is a credit against the next list rather than a write-off. The free Wear file still stands.
151. Struck — founded on a mistake, kept rather than deleted this archive
London Metropolitan Archives — STRUCK: the premise was wrong, FindMyPast covers London well
STRUCK WITHIN THE HOUR OF BEING RAISED, AND THE REASON IS THE POINT. This row was created on the finding that only one of 1,131 Downing baptisms for 1800–1806 is in London, read as a coverage failure. The control was wrong. Counting one surname by county tests where the Downings were, not whether the index reaches London. Of 100 SMITH baptisms in the same window, thirty-seven are London, Middlesex or Westminster. The coverage is excellent, no class of negative is invalidated, and nothing needs re-checking. The missing baptism is a real negative and it now argues that the family was not a London family — see the corrected row on the Welcombe marriage.
152. Done — finished and published this archive
I said four archives were ready for a second map layer. I had counted files, not read them.
The kit's Atlas gained a `schemes` prop on 20 September so an archive can ask more than one question of the same ground, and the case made for it was that D'Arcy, Blažević, Falco and Mazza already hold the data. They do not. Read properly: D'Arcy's 89 graves are keyed to *cemeteries with plot references* — “Lutwyche Cemetery, Brisbane (Port. GP5, Sect. 42A, Grave 78)” — and 18 of them match an atlas place; Blažević 33 of 72; Mazza 0 of 10. The coverage files are not geographic at all: Blažević's is per-person (ahnentafel, read/unread), D'Arcy's is per-database (FreeREG, Bristol, verdict *unusable*), Falco's is per-document (87 register acts), and Lerena's *film map* is a map of a film reel, thumbnail luminance against a threshold of 60, with no place in it. Nothing here is a place layer without a cemetery-to-town mapping or a geocode, and that is each archive's editorial work, not something to derive from outside. The mechanism is ready and unused; what each archive needs to supply is `cats: {scheme: cat}` on its own atlas rows. The claim was made from a file listing and should have been made from the files.
Closed 21 September: the recording was the deliverable. This row has no action left in it — it exists so that a claim I made to the estate, and which was wrong, is written down beside the numbers that disprove it. The mechanism it describes is shipped and the four archives know what they would have to supply. Leaving a correction open as though it were work makes the list longer and the record no truer.
153. Next — ready to start the Defranceschi session
Taking Defranceschi's map off its MutationObserver is a rewrite, not a removal
Offered as a next step and withdrawn on reading the component. `AtlasTimelineTint.astro` is 1,220 lines and the scheme switch is the smallest thing in it. It also carries: the panel's coverage-timeline tinting (gap rows, soft gaps, films, records, folds); `startOff`, which clicks the kit's own buttons after its first `apply()` because the kit keeps its filter state in a closure; the whole view in the address bar — `?layers=people,graves&base=terrain&q=Senj`, so a view is a link somebody can send; a basemap switcher; full screen; and a responsive rebuild of the controls. Ripping out the scheme half would remove a fifth of it and put the other four fifths at risk for no gain, because the kit cannot yet do URL state, basemaps or full screen. The sequence is the other way round: the kit absorbs those three — all of which the Record Atlas has independently invented too — and the local component then shrinks to the panel tinting that is genuinely this archive's.
154. Next — ready to start David, and the Geneology Map session
The Record Atlas already holds every place the family shelf layer holds — all 1,239 of them
Measured 20 September. The Record Atlas (`TheMap`, the standalone product) holds 9,472 places and 3,504 collections, with every name a place has ever had, who holds the paper, what it costs to look, and whose empire filed it. Defranceschi's map carries a *shelf* layer of 1,245 parishes — how far this family has got with the registers. 1,239 of those 1,245, or 100%, are already in the Record Atlas. Its other layers are not: people 72 of 165, graves 18 of 64, records 16 of 31, because those are about a family and the shelf is about records. That is the division of labour, and it is already true in the data. The global question — where are the records, who holds them, can I see anything tonight — belongs to the Record Atlas; a family archive should ask where it can put a named person, where the graves are, and what it has read, and link out for the shelf rather than keep a second copy of it. The Record Atlas's own component says it descends from the family Atlas and names three things it does better: colour by what a stranger can open rather than by how far one researcher has got, a year that filters rather than decorates, and a panel that fetches so the index stays 145 KB instead of 2.5 MB. It is the most evolved of the three maps in this estate and it is the one nothing else shares code with.
155. Next — ready to start David
Ask the bank about Wise — Anne has softened on it and it would be cheaper than SWIFT
Her invoices say REMITLY, WISE & PAYPAL ARE NOT ACCEPTED, and she has now explained why and partly withdrawn it. The problem is a five-page “mandate”: clients pay *anne brest photography*, and Wise wants the account in the name *AP Lapedus Brest*, so it fails and has to be cancelled and resent. But: *“someone paid me Wise the other day and I didn’t have to sign the Mandate, it came straight through to my account… maybe you need to ask your Bank. I was amazed.”* Worth testing before the next payment. A SWIFT transfer from Australia carries a bank fee plus an FX margin of three to five per cent; Wise would be materially cheaper and faster, and the transfer time was the thing that started this whole mess. If it works, say so — it solves a recurring problem for her other overseas clients too.
156. Done — finished and published this archive
End-to-end audit of the whole site — done, and it found three faults
Ran the pipeline clean from an empty dist, every gate, every JSON parsed, every tool compiled, and the checks nothing wires in. RESULT: 799 pages, 2,559 ids, all six site checks and all four repo gates green; 26 data files valid JSON; 14 Python tools and 2 Node tools compile; people.js and people.json agree at 350. THREE REAL FAULTS, none of which any existing gate caught. (1) A correction carrying cause `reasoning`, which existed in neither `causes` nor the page’s `label` map — live since 14 September, rendering with an undefined heading and an empty explanation. Cause added to both; check 14 added. (2) `namefold-extra.json` read by nothing at all — its 38 folds had been merged into namefold.json (verified: all 38 present, none differing), so no reader lost anything, but an unread live-looking data file is the Mazza trap with the catch off. Marked superseded; check 15 added. (3) /sources/ published that FamilySearch was down, against 7,276 citations. Fixed in the tool and the data. Both new gates verified in both directions. Check 15’s first version passed while the fault was in front of it, because its own comment named the file it was looking for and therefore always matched — caught by testing it rather than trusting it.
157. Done — finished and published this archive
No type checking on 41 .astro files — `astro check` is not installed
The only real gap the audit could not close itself. 36 pages, 4 components and 1 layout, and nothing type-checks any of them. `npx astro check` offers to install `@astrojs/check` and `typescript`, and `devDependencies` is currently empty — so this is a decision about adding two dev dependencies to the project, which is David’s and not mine to take. Why it is worth taking: every gate on this site checks DATA. Nothing checks the templates that render it. A mistyped prop or a renamed field would still build, and would fail silently in exactly the way this archive has now been caught by three times in four days. One `npm i -D @astrojs/check typescript` and one line in the build script.
Done 20 September, and it was right to be worth taking. `@astrojs/check`, `typescript` and `@types/node` installed, a `tsconfig.json` added (`moduleResolution: bundler`, which alone cleared 43 of the 102 first-run errors as a config artefact rather than a fault), and `check:templates` wired into the build. Four real faults, none of which the build had ever mentioned. Three `<blockquote>` opened and closed with `</p>` — two in documents.astro, one in method.astro — and one of them swallowed an `<h3>` and the paragraph after it: 928 characters of text inside a quote box meant to hold about 450. And an empty `<section class="blk">` opened at the end of people.astro and never closed, shipping `<section></section>` into every build since it was written. The gate is structural only. `astro check` reports 42 more things here and every one is TypeScript not knowing the shape of an imported JSON — *Property 'log' does not exist on type 'unknown'* — which are not faults, and a gate failing on them would be switched off within a week. It fails on a tag that does not close or a parse error, counts the rest, and says the count. Proved by reopening a blockquote: exit 1 naming the line and the end of file it ran to; restored, exit 0.
158. Struck — founded on a mistake, kept rather than deleted this archive
A correctly-placed pin with nobody on it is worse than a pin in the wrong country
Raised by the Defranceschi session on 20 September, out of fixing the eight misplaced pins the kit's new `checkplaces.py` found. Pinning a place by hand only keeps its people if the pinned name keeps the same `head()` — the part before the first comma — as the string the roster attaches on. So *Dolo, Venezia* works and *Dolo Veneto* would not have: the pin lands in the right field in Italy and every person silently falls off it. That failure is worse than the one just fixed, because a wrong pin looks wrong to anybody who knows the geography and an empty right pin looks perfect. They caught it by checking counts after every rebuild — 3, 2, 1, 1, 1, all retained — which is a discipline and not a gate. The gate-shaped version is a baseline diff: refuse when a rebuild drops a place's attached people. Not built, because it needs a stored baseline and a decision about what a legitimate drop looks like, and that is design work with the asking archive absent. Recorded so it is not lost.
Reframed by that session, 20 September, and part-measured here. Not per-place but per-person: refuse when a person who had a place stops having one. That side has no legitimate drop they could think of, and it sidesteps the stored-baseline question because it compares the two sides of one rebuild. Measured on this archive, which is the bare-names case that made the country gate inert: 65 places name 86 distinct people, all 86 held here — 25% of the 347 — so per-person has purchase where per-country had none. But there is a legitimate drop, and it is here. The link is by bare name and two of those 86 names are shared: George Mountjoy, and Petrus Jacobus Booysen, who is three men on one page. Every place naming him attaches to all three, and the day this archive disambiguates that, two people lose every place they had — which is the archive getting more careful, not less. Proposed narrowing: refuse when a person whose name is unambiguous here loses their last place. That watches 84 of the 86 and leaves the namesakes to a human. Whether it survives their three cases — a merged duplicate place, a corrected roster row, a resolved household — is theirs to measure. Not built: two archives is a sample of two, and a rule needing a fourth exception is not a rule.
STRUCK 20 September, and the reason is worth more than the gate would have been. Three versions were proposed and each was killed by a different archive's data. Per-place died on legitimate drops: merging a duplicate place, correcting a roster row, resolving a household all move people off a place lawfully. Per-person died on namesakes — and not on mine. This archive has 2 shared names among 86 placed people, so a narrowing to *unambiguous names only* would have watched 84 of 86; Defranceschi has 103 shared names covering 378 of 1,225 roster rows, so the same narrowing would watch 706 of 1,057 and go blind on the 31% where that archive is doing its hardest thinking. Their point is the sharp one: *unambiguous* is not a property of their data, it is a property of a resolution step they have deliberately not taken. Their *Maria Defranceschi* is 24 rows nobody has decided anything about; my three Petrus Jacobus Booysens are three men this archive knows are three men. Per-record — their proposal, the unit that cannot split — died here: 7,274 of 7,624 register rows carry an ark, so identity is fine, but a record in this archive carries no place at all; the 1,017 rows that name one do it in free prose, findable only by the string matching that put Pag in Mindanao. The finding is not about gate design. Booyzen attaches a place to a person by bare name; Defranceschi attaches it to a roster row; Booyzen's records attach it to nothing. There is no shared unit of attachment in this estate, so any gate watches whichever archive it was written in. If a place id ever lands on a record rather than a place name in a sentence, this becomes trivial and can be un-struck.
The sharper version, from the Defranceschi session, and the part to remember when reaching for a gate: *the gate was always going to be written in whichever archive wrote it, and would have looked correct there.* Mine would have watched 84 of 86 and reported green here while blind to a third of their roster. Theirs would have watched arks and found nothing to attach them to in this one. Both would have passed their own tests. That is the failure mode, and it is not specific to this gate.
159. Done — finished and published this archive
Nothing had ever checked the markup, and seven of the eight archives were shipping broken tags
`check:templates` went into all eight builds on 20 September and found nineteen real faults across seven archives. None had ever been mentioned by a build: Astro closes a tag it does not understand, ships, and the page looks nearly right.
The two that changed what a reader sees. *D'Arcy /places/* shipped two search boxes — the prominent one, under a heading reading *Search* with a placeholder about Boulia and Staffordshire, drove a script querying `.prow` and `.plgroup`, of which the page rendered none. It did nothing, and the count beside it read *167 places* and could never read anything else. Left behind when the page moved onto the kit's `Places` component; 49 lines of dead search UI, dead script and dead CSS removed. *Lerena /direct-line/* carried `href` set to the quoted string `{u(…/enrolment-card/…)}` rather than the expression — quoted, so a literal string and not an expression — which shipped as the literal opening of that quoted string and made a dead link to that archive's key document. The link gate never saw it because it is not a path.
The rest: three `<blockquote>` closed with `</p>` here, one of them swallowing an `<h3>` and the paragraph after it — 928 characters inside a quote box meant to hold 450; an empty `<section>` shipping in every Booyzen build; a `<b>` nested in a `<b>` and never closed on Lerena's home page; nested `<p>` in Falco; orphan `</tbody></table></Scroll></section>` in two archives where a pre-kit table had been; and five doubled or stray closing tags. Mazza was the only archive with none.
The gate is structural only. `astro check` also reports 10–124 findings per archive that are all TypeScript not knowing the shape of an imported JSON, and a gate failing on those is a gate switched off within a week. It fails on a tag that does not close, counts the rest, prints the count. Proved by reopening a blockquote: exit 1. [21 Sep, edited by another session: two literal link-attribute fragments in this note were being parsed out of the rendered page by links.py and reported as broken links on /worklist/ — THE NOTE DESCRIBING THE BUG REPRODUCED IT, and the first attempt to fix it reproduced it a third time. Rewritten to keep the meaning with no attribute syntax in it at all.]
160. Next — ready to start David
Commits keep sweeping up other sessions' work — four times into mine, and once out of it
Not a fault in anyone's work and nothing is lost, but the record is wrong and it keeps happening. On 17 September the David's Map change was committed by the Defranceschi session inside *“All four godparent sets read”*; on 20 September the template fixes for Falco, Mazza and Lerena went in under *“Michele and Michelangelo Falco are two men”*, *“A second brother, and eight Arena grooms”* and *“Fourteen records name him”*. Each of those messages describes work that is not the change it carries. The cause is `git add -A` in a repo two sessions are editing, and the effect is that `git log -S` finds a fix under a message about something else — which matters in an estate whose commit messages are the change log. A session could stage paths instead of everything, which is what this archive does; whether to ask that of every session is David's call, not mine to impose across eight repos. 21 SEPTEMBER — AND IT WENT THE OTHER WAY, WHICH MAKES THIS A TWO-WAY RECORD RATHER THAN A COMPLAINT. Commit d3828d3, “The tailor was right: it is James Hogg”, 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. Its work went in under a message describing only the Mirror story. What was actually in there besides the tailor: General Willem Johannes Kolbe (row 140 — the rank is not an honorific: born 9 July 1847 in the district of Philippolis, led the Bloemfontein commando, made general at the siege of Kimberley, at Paardeberg, Driefontein and the Brandwaterkom, in De Wet’s Cape invasion of February 1901, two horses shot under him, died 1941, and already held here as a line in the WO 108/368 prisoner list); row 170, the *Herinneringe van genl. W J Kolbe* memoir at VAB; row 137, that the DSAB was terminated after Volume V in 1987 so Schoeman’s 1992 draft was written for a series that had already stopped; rows 171 and 172; and the `#pow-wo108` rename across people.js, searched.json and dossiers.json. NOT AMENDED. The commit is pushed and two sessions are working in one tree; force-pushing to rewrite shared history would be a worse fault than the one it fixed, and this archive records its mistakes rather than erasing them. The correction below carries the attribution instead. The rule that was broken is this archive’s own: stage named paths, never `-A`, not even scoped.
161. Done — finished and published this archive
check:living read only HTML, and an atlas serves its people as JSON
Raised by the Luwinski session, which has two children flagged living in its emigration data and said plainly that the gate guards the built pages and not that JSON. It was right and the gate walked past every `.json` in `dist` — an atlas fetches its places from one, a search box its index, and both are served to anybody with the URL. A living person could sit in `atlas-data.json` with a date and a pin and the build would pass, because the name never reaches a page as text; it reaches the browser and is drawn. Measured before changing anything, and the first measurement was wrong. Three archives appeared to carry living names in served JSON — Falco 97, Mazza 424, Defranceschi 99 — which read like a serious leak. Every one is also the name of a published dead person: a namesake, which the gate's own `published_names()` already subtracts. Living names in served JSON belonging to no published dead person, across the estate: zero. The blind spot was real and the leak was not. Proved on a scratch archive: a living person's birth date in a served `atlas-data.json` and none in the HTML — the old gate reports *ok*, the new one names the file, the date and the person. A first falsification used a bare name and the gate passed, correctly: under `named-bare` the name alone is not the fault, the date is.
162. Done — finished and published this archive
Two fixes for the fold-target bug were worse than the bug
The Luwinski session found that `namefold.py` chose a cluster's target by frequency, and frequency tracks how often an archive paraphrases, not which document is closest to the event — their Hostovsky is on the 1956 marriage register, the most formal of three documents, and is not the commonest form. First fix: refuse to fold any cluster the archive spells more than one way. That dropped Booijsen / Booijzen / Booysen / Booyzen / Booÿsen in this archive — five real people's surnames, and the one fold this estate most needs. Second fix: choose the target alphabetically. Worse, and subtly: the safe/proposed split is computed relative to the target, so Booijsen as target makes Booyzen a letter-change away and it stops folding — 1,495 register rows would have gone back to 59. Caught by regenerating the map and reading it, not by reasoning about it. What shipped: the selection is unchanged and the *output* is what changed — the tool now says which label it used and that it is only a label. The bug was the implication, not the choice. Also found on the way: this archive's own map was generated once by hand and regenerated by nothing, so it is a build step now.
163. Done — finished and published this archive
The name clusterer was reading the work list as if it were a register
Found by the Blažević session. `namefold.py`'s `corpus()` scanned every `*.json` in the data directory, which in this estate means the work list, the searched register, the corrections and the change log — the archive's narrative. A surname appearing only in a *sentence about* a surname entered the corpus with the standing of a register row. Their own instance is mild: *“Subrinic, Xubrinich and Zubrinicz are zero”* is a sentence about a failed search and its two tokens are nobody's name. The dangerous instance is the example this tool already carried — an archive whose work list says *“the other Roos family, who are not ours”* was handing that name to the clusterer with a record's confidence, and a fold that joins two families is the worst thing the kit could ship. What had saved it was the counts: 2 tokens against 1,059 announces itself. That is a coincidence of scale, not a guarantee. Sixteen narrative filenames are skipped now; Subrinic leaves Blažević's clusters and this archive is unchanged at 13 and 70.
164. Done — finished and published this archive
A number quoted to another session needs a timestamp, because their data moves under it
I told the Falco session their waiting fold was *Sansone → Sanzone, 312 tokens, typing SANZONE reaches 15%*. They replied that it is Sanzone → Sansone, 24 tokens, 94% and asked what in my rollup was reading the wrong side of the pair. Nothing was. When I measured, their `people.json` held Sanzone as the archive's own spelling and the anchor followed it; by the time they read my message they had **re-read the initial as a long ſ against the *z* in Crescenzo in the same hand, corrected the file, and the anchor had followed that instead. The tool did exactly what it should. I quoted a reading that had been superseded and did not say when I took it.** Three of my figures went stale the same way tonight — the estate is eight sessions correcting each other's data while messages are in flight, and a bare number in that traffic reads as a current claim. Say when it was measured.
165. Done — finished and published this archive
A withdrawn reading stated as fact now refuses the build — and a dated log is not a claim
Designed in the Falco archive and ported to the kit with their permission as `checkretired.py`. This estate has tried a general prose check four times and abandoned all four at roughly 100% false positives. This one works, and Falco's one-sentence reason is why: *a retired phrase is the archive's own wording, never a document's, so an unquoted hit is a claim and a quoted one is a quotation.* «act 26 of 15 December 1814» is not something a register says. The same trick cannot be used on a retired surname, because the documents contain the surname. The withdrawal mark is a parameter, not guillemets, so an archive using `<del>` or a *was:* field is not fired on for an honest withdrawal. Three real withdrawn readings declared here — the NAAIRS URL with a dot that has no address record, “Leeuwberg pinned at Fauresmith”, and “buried at Volksrust”. It found four hits on its first run and every one was correct to find, and wrong to DELETE — which is not the same as nothing to do. All four are in `/searched/` and the work list, which record what was asked on a date and what was believed when a row was written. This archive still says Martha Angeline Kolbe was “buried at Volksrust”; it withdrew that on 14 September when her stone was found at Wakkerstroom Main Cemetery. That is not a stale claim — it is an accurate record of a reading held that week, and a log edited whenever the belief changes is not a log, for the same reason a struck row is struck and not deleted. Those two files are named as exempt, with that reason, in `retired.json`. Everything else is checked. Proved both ways: the phrase stated as fact in `places.json` exits 1 naming the file, the date and the reason; the same words inside the archive's own quotation mark exit 0.
The carve-out lasted an hour and the Falco session was right to kill it. I had exempted `/searched/` and the work list on the ground that a dated log records what was believed then. Their test is better: **not *is this a log* but *would a reader act on this row today* — and a searched register is acted on precisely so nobody repeats a search, so exempting it removes the check from the page that most needs it. The third option is the one both archives already use elsewhere: keep the withdrawn text and put the correction in the same row.** All five rows now read *she was “buried at Volksrust” — withdrawn on 14 September, her stone is in Wakkerstroom Main Cemetery*; the history stays, the reader cannot act on the dead half, and no file is exempt. They also caught me writing *every one was correct to find and wrong to fix*: correct to find and wrong to delete is not the same as nothing to do, and the elision is how a carve-out gets justified. And the gate caught me twice more while I wrote this row — naming the retired phrases in italics rather than in the mark, which is exactly the failure Falco predicted: the person most likely to trip it is the person writing the correction.
166. Done — finished and published this archive
The test is not whether two paths disagree — it is whether a reader can stop early and be wrong
The most transferable thing to come out of 20–21 September, and it is the Blažević session's sentence, not mine. I had written that two independent paths to a conclusion, one broken and the working one hiding it, was the shape behind the week's faults — the atlas drawing a spouse from `kin.js` while `marriages.json` reached no page; a search box driving a dead script beside a filter that worked; a fold map built correctly and read by nothing. They corrected it: the covering path can be the same path, earlier in the file. Their three archive notes answered a question wrongly at the top and correctly at the foot, so every tool that read the file found something true — recency covered for the error the way a second code path would. So the test is not *are there two paths* but can a reader stop early and be wrong. That is also true of both of mine: the atlas panel and the dead search box each let somebody stop at the first plausible answer. Their structural response is the part worth copying: rather than detect that one sentence contradicts a later one — which a build should not attempt — insist the reader meets the withdrawal FIRST. Cheap, always satisfiable, and therefore a gate rather than a report. `checkretired.py` already covers their specific case, because an unmarked withdrawn phrase is a claim wherever it sits in the file; what it does not yet do is notice that a *marked* one is a screen above its correction.
167. Done — finished and published this archive
Two pages promised two different living-person rules, and the stricter one was unchecked
Prompted by the D'Arcy session, whose `/about` — the page that asks a reader to take the rest on trust — said living people are *“not published at all. Not hidden, not gated — absent from the build”*, while six are named deliberately and had been listed honestly in their own `living.json` the whole time. The sentence was written from the rule and never read against the data beside it, and their gate could not catch it because it runs under the policy that permits exactly what the six are. They called it an overclaim *in the direction of more rigour, which reads as reassurance, so nobody checks it*, and asked the estate to read its own sentence against its own data. Here the two sentences disagreed with each other. `/register/` said *“the living are not here, as everywhere else in this archive”* — absence — while `/about/` says *living people are named and nothing more*, which permits presence. Both cannot describe one regime. Measured: 350 people, none flagged living, and the single person born inside eighty years has a death recorded. So the register genuinely holds no living person — which is a fact about this family and not a policy, and the sentence now says which, states the estate rule where it would apply, and adds that here it has not had to be applied at all.
And the first fix was too small, which is the transferable half. The D'Arcy session went back after correcting its `/about` and found the same claim in its site footer, on all 874 pages — the single most-published sentence in that archive and the one nobody had ever re-read, in a layout file it had been treating as off-limits. Their lesson: correcting the instance you were shown is not the same as correcting the claim; a proposition repeats itself in several wordings and one of them is usually in a layout. Swept here on that prompt — my own first pass had searched `pages` and `data` and not `layouts` or `components`. The footer on 797 pages turned out to state the rule correctly and to add *“this archive currently holds none”*, which was true. But it was asserted, and `/about` promises that *what it holds is counted from its own data, not asserted* — so the footer was the exception to the archive's own promise, and the day somebody adds a living person it would have become false everywhere at once in the place nobody re-reads. It is counted now. Proved by injecting a living row and rebuilding: the footer reads *“One person here is living”*, and says *“This archive holds none”* with the row removed.
Third pass, and I committed the fault I was writing about. The sentence that replaced the register's overclaim said *“of 350 people it holds, not one is living”* — a typed cardinal, correct when written, in prose written specifically to fix an asserted claim. The D'Arcy session did the identical thing in the identical hour, typing *“six are named on purpose”* into its own corrected footer, one hour after recording that six of its thirty-six own errors are numbers typed instead of counted. Both counted now. Their sequence is the lesson and it is worth keeping whole: correct the instance, then discover the claim, then discover that the correction asserted what it should have counted — three passes, each of which felt like the finish. Also checked here on their prompt about template expressions: `/about`'s *“one of five words”* is correct, five levels are listed on `/method` — documented, probable, inferred, family, disputed — so that one is a typed number that happens to be right and stays right, because the vocabulary is the estate's and not this archive's count.
168. Done — finished and published this archive
The living count was right about a list, and the list was missing the living
The D'Arcy session's last test, and it found more here than the typed numbers did. Their rule: *when a typed cardinal describes your own data, the number is the symptom — ask what the number counts and whether the thing it counts is complete.* Theirs was a spellings table accurate at nineteen rows and missing SANIGOR, the spelling the subject of their own error 28 was baptised under. Applied here, the footer's *“this archive holds none”* counted `people.json` and was wrong twice.
One. *Tersia Booyzen* — whose testimony this archive quotes in full on `/rebels/` — carries `"b": "living"`, the word in the birth field rather than a flag, so `checkliving` never counted her and every gate reported zero. She is flagged properly now and the footer reads *One person here is living*. No dates or places were published for her, so nothing leaked; the count was simply blind.
Two. The home page says *“This archive was built for David Defranceski, born in Johannesburg, living in Brisbane”* — two places for a living person, which is precisely what the footer's own rule forbids. It is his archive, his brief and his own name, so the answer is not to delete it but to stop the rule claiming something the site does not do. The exception is stated in the footer now. That is the D'Arcy fault arrived at from the other end: theirs promised absence while six were named deliberately and recorded honestly; mine forbade places while giving two, by choice. An exception that is real and unstated is the overclaim.
169. Done — finished and published this archive
Read the images behind the two Kolbe–McCabe marriages, not just the index
Index entries found 21 September; the images are not read. *South Africa, Civil Marriage Records, 1801–1974* — Anna Jacoba Kolbe × Francis McCabe, 13 July 1871, and Jan Hendrik Kolbe × Rose Maria McCabe, Bloemfontein, 4 June 1872. A civil marriage entry gives ages, occupations, fathers’ names and witnesses, and the fathers are exactly what is wanted: they would show, from a register, whether Jan Hendrik is the Commandant’s son and whether Rose Maria and Maria Petronella are Francis McCabe’s daughters — the one link currently borrowed from a user tree. Both should be readable as images on FamilySearch. 21 SEPTEMBER — THE INDEX IS EXHAUSTED AND IT DOES NOT CARRY THE FATHERS. The record page for the 1872 marriage (FamilySearch ark 1:1:6ZHB-XQ1R, image ark 3:1:3Q9M-CSK8-G98V-M, DGS 008145483) gives: JAN HENDRIK KOLBE, aged 23, birth year 1849; ROSE MARIA McCABE, aged 17, born 1855; married 4 June 1872 at Bloemfontein. Pulled through the archive’s own documented image-data endpoint, which returns the indexed GEDCOM-X: the record holds the couple and nothing else — no parents are indexed. SO THE FATHERS ARE ONLY ON THE IMAGE, and reading it needs eyes on handwriting rather than an endpoint. The image exists and is linked from the record page. AND THE AGES ARE A GAIN ON THEIR OWN. Rose Maria McCabe was seventeen, and she married four months after her father Francis McCabe died on 23 February 1872. A fifth bride in this archive married before eighteen. Jan Hendrik at 23, born 1849, sits correctly among the Commandant’s children — the Commandant married in 1843. 27 SEPTEMBER — BOTH IMAGES READ, AND THE ROW'S OWN PREMISE WAS WRONG. This row said a civil marriage entry gives fathers' names. The Free State form of 1869–72 does not have the columns, so the fathers were never obtainable here and the row could not have succeeded as written; that is logged at /corrections/. WHAT THE PAGES DID GIVE: the 1871 entry says WEDUWE and “KOLBE, GEB: VAN DER MERWE”, which settles the duplicate on a document; and both entries put these people on LEEUWBERG, the Commandant's farm — the widow in 1871, and Jan Hendrik, veeboer, with his bride in 1872. A G. A. KOLBE WITNESSED THE 1872 MARRIAGE. The one link still borrowed from a user tree — that Jan Hendrik is the Commandant's son — is not closed, and VAB MHG M777 remains the document that would close it.
170. Next — ready to start Free State Archives, Bloemfontein
“Herinneringe van genl. W J Kolbe” — the General wrote his own account, and it is in Bloemfontein
A MEMOIR BY A MAN OF THIS FAMILY EXISTS AND NOBODY HERE KNEW IT. Found 21 September in a footnote — **Fransjohan Pretorius, “Straf op Kommando gedurende die Anglo-Boereoorlog, 1899–1902”, *Military History Journal* Vol. 8 No. 3, which cites VAB, A1 19, Renier-versameling, No. 244, *Herinneringe van genl. W J Kolbe*, p. 78. (The journal prints the reference as “Al 19”; read as A1 19.) The citation is to page 78, so it is not a note — it is a document of some length. Pretorius is the historian of Boer commando life and does not cite loosely. WHY IT MATTERS MORE THAN ANOTHER REGISTER ENTRY. This archive holds no word written by anybody of this family between the Settler’s death in 1844 and the twentieth century. If these are the reminiscences of Willem Johannes Kolbe (1847–1941) — the General of row 140, born in the district of Philippolis, grandson of the Settler — then they are a grandson of the Settler describing his own life in his own words, and they would settle from the inside the descent that row 140 still holds at arm’s length. They may also name his brothers and sisters, which is the twelve-children question this archive has been unable to answer since the Commandant’s obituary was read. IT IS AT THE SAME REPOSITORY AS THE ESTATE FILE ALREADY WANTED. VAB is the Free State Archives Repository in Bloemfontein, which already holds VAB MHG M777 (the widow’s 1896 estate, filed under McCabe) and VAB MHG K71 (the Commandant’s own, 1868). Three references, one repository, one enquiry — this row should be ordered with those, not separately. CAUTION, AND IT IS A REAL ONE.** The identification rests on initials. A National Archives node for a *Renier-versameling* in Pretoria — a different repository — lists genealogical correspondence with a “P Kolbe”, so the collector corresponded with more than one Kolbe and the collection is split across depots. W J Kolbe is not proved to be Willem Johannes Kolbe of Panfontein until somebody opens the folder. What is certain is that it exists, where it is, and that it is worth the asking. 21 SEPTEMBER — THE LETTER IS WRITTEN, AND IT CARRIES ALL THREE, at `requests/free-state-archives-kolbe-three.md`. One enquiry, not three, exactly as this row asks: the memoir first, then VAB MHG M777 and VAB MHG K71, with the death notice named as the page most wanted from each if the files are impractical. The caution in this row is in the letter itself — it tells the repository that the identification rests on initials, that a *Renier-versameling* in Pretoria lists a *P Kolbe*, and asks to be corrected if `A1 19` is a misreading of the journal's printed *Al 19*. It also asks rather than assumes about the four further VAB references, on the principle that a short list is answered sooner. Address confirmed from the Free State department's own page; a second address in wide circulation differs from it by one letter and is recorded as unconfirmed.
171. Done — finished and published the kit and estate
Two sections on the documents page had the same id, and fourteen links went to the wrong one
FOUND WHILE LINKING A NEW RECORD, WHICH IS THE ONLY REASON IT WAS FOUND AT ALL. The documents page carried two `<h2 id="pow-1901">`: *1901 · Three Booyzens on the* City of Cambridge, and three thousand lines later *1900–1901 · Six Kolbes and three David Schalk Booysens*, the WO 108/368 list. Fourteen links meant the second and every one of them landed on the first — five people records, five dossier entries, two lines of *What Was Searched*, and work-list row 90, all labelled *The prisoners* and all arriving at a ship's manifest. The contents list carried both sections and pointed both entries at the same place. The second heading is now `#pow-wo108` and the fourteen are repointed; the five links that genuinely mean the *City of Cambridge* were left where they were, identified by their own label rather than by guesswork. THE GATE COULD NOT SEE IT, AND THAT IS THE MORE USEFUL HALF. `checkarchive` collected every id on a page into a set and then asked whether the fragment was in it. A duplicate answers yes. The check was built to catch a fragment pointing at a section that had been renamed, and it did that perfectly for a year while being structurally incapable of seeing two sections answering to one name. Counting instead of collecting now reports it, and the kit is repinned here at `6dfd7d7`. IT ONLY REPORTS A DUPLICATE THAT SOMETHING LINKS TO, AND THAT WAS MEASURED BEFORE IT WAS WRITTEN. Across the seven built archives, failing on any repeated id would have failed Defranceschi on thirty-six findings of which one mattered — the rest are ids nobody links to, invalid HTML that harms no reader. Failing only on the linked ones finds this archive's `pow-1901` and Defranceschi's `/diaspora/#argentina` and is silent on Blazevic, D'Arcy, Lerena, Mazza, Luwinski and Our Family. A gate that cries wolf is worse than no gate, so the scope was chosen from the estate's actual data and not from what would have been tidier to write. The list
172. Next — ready to start the Defranceschi session
Defranceschi's diaspora page has two #argentina sections, and two links reach only the first
FOUND BY THE NEW KIT CHECK, NOT BY READING THAT ARCHIVE, and it is that archive's to fix — nothing there has been touched from here. `/TheDefranceski/diaspora/` carries `id="argentina"` twice: once on the prose section that ends *“Argentina thin in the indexes, thick in the tree from the 1880s — Buenos Aires”*, and once on the later panel headed “Argentina · The Argentine pen, opened”, about the placeholder node in the live tree. Two links point at the fragment — from `/searched/` and from `/worklist/` — and both can only ever reach the first. Judging by their context, the second is the one they mean. The fix is the same shape as this archive's: give the later section its own id and repoint whichever links mean it, leaving the rest. Do not take the id from the heading text alone — that is how two sections came to share one in the first place. Defranceschi also has thirty-five other repeated ids that nothing links to; those are invalid HTML and break nothing, and the gate is deliberately silent on them, so this row is one anchor and not a cleanup. The kit is at `6dfd7d7`; that session will see this the next time it repins.
173. Next — ready to start David
Somebody published a descendants study of Frederik Fortunatus Kolbe in 2008, and its own numbering disagrees with this archive
A PUBLISHED GENEALOGY OF THIS EXACT BRANCH EXISTS AND THIS ARCHIVE HAS NEVER CITED IT. **Van Aardt, Pieter, “The Descendants of Frederik Fortunatus Kolbe (b2c8); a branch of the Kolbe family”, *Familia* 45(4), 2008, pp. 247–250.** Found in the eGGSA index to *Familia*, the journal of the Genealogical Society of South Africa. Four pages, peer-published, and about the one name this archive traces across three families. AND THE CODE IN THE TITLE IS A TEST, WHICH IS WHY THIS IS NOT JUST A REFERENCE TO FETCH. `b2c8` is de Villiers/Pama: the eighth child of the second child of the progenitor a1 — that is, a grandson of whoever Van Aardt takes as the stamvader. This archive holds Frederik Fortunatus Kolbe (b. 1839 Bethulie, d. 23 Nov 1929 Utrecht) as the SETTLER'S TENTH CHILD — on that reading he is b10, a son, not b2c8. One of three things is true: Van Aardt's a1 is not the 1820 Settler; or his Frederik Fortunatus is a different man of the name; or this archive's birth order is wrong. All three are worth knowing and the article settles which. Row 140 already warns that there are at least four men called Frederik Fortunatus Kolbe, and a numbering that disagrees with ours is exactly the kind of collision this archive has withdrawn a claim over before. A SECOND NAME TO WATCH, NOT TO ADOPT. A Frederik Fortunatus Hofmeyr Kolbe (27 Sep 1919 – 18 Jun 1969) of WAKKERSTROOM, Transvaal turns up in tree indexes — Wakkerstroom is this archive's Kolbe place, and the Fortunatus name is this archive's name. Nothing is claimed: he is from a user tree, and the resemblance is the whole of the evidence. He is noted here because the same string also appears in the DSAB's printed index (see row 137), where it is probably a misreading, and the two should not be quietly merged into one man. WHERE TO GET IT, AND WHY THIS ROW IS NOW DAVID'S. *Familia* is the GSSA's quarterly. eGGSA indexes every article and publishes none of the text. There are two ways to the four pages and both are a transaction: individual back numbers are sold through the eGGSA online shop, or **GSSA membership includes *Familia* back issues as PDFs. Neither is this archive's to do** — buying and joining are David's, and the second is worth weighing on its own merits against row 12's gendatabase question, because a membership that carries the whole run of *Familia* reaches far more than one article. It is still the cheapest substantial source outstanding on this list — four pages, against archive orders and a trip to Bloemfontein. Nothing further can be done here until that is decided.
174. Done — finished and published this archive
I cited a commit hash to another session without reading it, and it does not exist
Small, and exactly the same shape as everything else logged today. Writing to the landing-site session about the attribution commit, this archive cited `6e4a33f`. There is no such object — `git cat-file -t` fails on it. The commit is `0d867c3`. The hash was typed from memory of having just made a commit, rather than read back out of `git log`. Nothing in the archive itself was wrong — row 160 and the corrections row both cite `d3828d3`, which is the commit that actually swept the work, and that is correct. The error was only in the message to a peer, who checked it and said so. THE RULE, IN THEIR WORDS AND WORTH KEEPING: a reference is code, and an untested reference is untested code. They had hit the identical fault the same day — the archive.org identifiers in row 132’s “method that worked” had a spurious character in them, so every id in the note returned nothing. Both were references written from memory in a note about how to find things.
175. Done — finished and published the kit and estate
Swept all eight archives for generators that read their own output — one had it, and seven are clean by construction
THIS ARCHIVE FOUND THE FAULT IN ITSELF AND THEN ASKED WHETHER IT WAS ALONE, because a fault found by accident in one place is not evidence about the other seven. The test, stated so it can be re-run: a tool is at risk only if it reads `site/dist` AND writes a file that is rendered back into `site/dist`. A checker that reads the built site and writes nothing cannot loop; a generator that writes something only *fetched* at runtime cannot loop either, because its output never becomes page text. THE RESULT. Tools that read `dist` and write data rendered back into it: `dossiers.py` and `gallery.py`, both here — and `gallery.py` already excludes `/gallery/`, so it was written correctly. The four search indexes — this archive, Blazevic, Mazza and Luwinski — all write JSON that the search page fetches at runtime and never inlines, so nothing they produce comes back as text. Falco's `check_release.py`, Lerena's seven `dist` readers and Blazevic's `check_site.py` write nothing at all. D'Arcy, Defranceschi, Our Family and the Record Atlas have no tool that reads `dist`. One archive in eight had it, and it was this one. A NEGATIVE I NEARLY REPORTED AND SHOULD NOT HAVE. The four search tools do not strip `<script>` before reading the HTML, and a grep for JavaScript tokens in the built indexes returned 25, 24 and 6 hits. Every one was the word “document.” in ordinary prose — *“a match, not a document.”* — caught by a pattern meant for `document.` A regex is an assertion about what a string means, and this one was wrong. The indexes are clean. WHAT THE ESTATE GETS OUT OF IT. Written up in the kit as *A generator that reads its own output* (`kit/docs/traps.md`, `6bace55`), with the three things worth more than the case: two runs of a generator are not a convergence test if both read the same `dist`; the visible symptom was the smaller half — stripping the quoted counts moved the churn only from 347 of 347 to 343 of 347; and the exclusion has to be a section, not a route, because a person page is real prose with a block of generated output inside it. The cheap standing check is one number: what the data file stores against what the built pages print. Here that is 9,008 against 8,972, and one page prints *more* than the file holds, which no rendering filter can do. It should be zero.
176. Done — finished and published the kit and estate
Four archives, four different arrangements for the search index, and only one of them is right
FOUND BY THE KIT GATE WHILE SWEEPING FOR SOMETHING ELSE, AND IT IS THAT ARCHIVE'S TO FIX — nothing there has been touched from here. `checkarchive` over Luwinski's built site: `FAIL searchindex — /search/ fetches /searchindex.json and it was not published`. Verified rather than taken from the gate: 27 pages built, no build in flight, and neither `site/dist/searchindex.json` nor `site/public/searchindex.json` exists. IT IS STRUCTURAL, NOT A ONE-OFF. `tools/build-search.py` writes to `DIST/searchindex.json`. The build chain does run it in the right order — `astro build && … && python3 ../tools/build-search.py` — but `astro build` empties `outDir` before refilling it, so the index exists only between the end of one build and the start of the next. Any bare `astro build`, any interrupted build, any rebuild that stops short of that step, and it is gone with nothing to notice. THE COMPARISON IS THE ARGUMENT. The three other archives that ship a search index — this one, Blazevic and Mazza — all write `site/public/searchindex.json`, and Astro copies `public/` into `dist` on every build, so theirs cannot be lost this way. Luwinski is the only archive writing into `dist` and the only one whose search is broken. Changing the output path should be the whole fix, and it makes the order of the build chain stop mattering. This is the failure `checkarchive`'s `searchindex` check was written for in the first place — search shipped broken on three archives at once, and the page loaded and found nothing. It has now caught a fourth. Also reported in passing, for that session's judgement and not this one's: non-canonical confidence words there — *not established*×3, *ruled out*×2 — against the estate's five. 21 SEPTEMBER, CORRECTED BY THE LUWINSKI SESSION ON TWO COUNTS, AND THE SECOND ONE CORRECTS THE ADVICE. FIRST — THE DEPLOYED SITE WAS NEVER BROKEN, AND THIS ROW SAID OTHERWISE. `https://daviddef.github.io/TheLuwinski/searchindex.json` returns 200 and 227,974 bytes, because CI runs the whole chain in order on a clean checkout. The fault was local-only, with a structural cause — which is worth fixing and is not the same as shipping broken search. An error in the direction of alarm is still an error, and it is corrected here rather than quietly edited away. SECOND — “JUST WRITE TO `public/`” WOULD HAVE TRADED A LOUD FAULT FOR A QUIET ONE. Astro copies `public/` into `dist` during the build. A generator that runs after `astro build` and writes only to `public/` therefore ships the previous build's index, permanently one generation stale and invisibly so. A 404 announces itself; a slightly wrong index does not. SO I CHECKED THE OTHER THREE RATHER THAN PASS THE ADVICE ON, AND ALL THREE DIFFER. MAZZA IS CORRECT AND PAYS FOR IT: its chain is `… astro build && npm run searchindex && astro build` — it runs astro twice, so the second pass copies the fresh `public/` into `dist`. Verified: the two copies are byte-identical, same mtime, different inodes. On a machine at load 214 a second full astro build is not a small price. BOOYZEN AND BLAZEVIC HAVE A THIRD PROBLEM: the generator is not in the build chain at all. Neither `build` script mentions it; it is run by hand and lags by however long since someone remembered. Measured here rather than asserted: 795 of 798 built pages are in the index, and the three absent — `/places`, `/search`, `/who` — are excluded on purpose. So no page is missing and the drift is content-level: text written at 07:05 is absent from an index built at 06:27. Milder than this row first implied, and still wrong. THE BEST ARRANGEMENT IS LUWINSKI'S: write both copies — `dist` fresh at once, `public` durable against an interrupted build — without Mazza's second astro pass. AND THE GATE HAS A GAP WORTH NAMING. Luwinski added `check-search.py`, which reads the `.astro` sources for what the page actually fetches and then checks that each file exists, parses and is non-empty — good, and it derives what to check from what the page asks for, so a new fetched JSON is covered without anyone remembering. But existence, parse and non-empty all pass on a stale index. Booyzen's has 12,064 rows, is perfectly valid, and is behind. Freshness is a different question, and the cheap form is the same shape as the dossiers check: compare the artefact against `dist`, not against itself. Sample the built pages, confirm each is represented, and a page built after the index was written shows up as one number that is zero or it isn't.
177. Done — finished and published this archive
Two names out of a compiled Kolbe report — a child not held here, and the great-uncle who would explain “Rector”
A FORTY-ONE PAGE REGISTER REPORT FOR THE LONDON TAILOR HAS BEEN SITTING ON A BARRY FAMILY WEBSITE SINCE 2012 and this archive had never opened it — *“Kolbe Family Tree”*, southafricanbarryfamilytree.wordpress.com, 29 December 2012, crediting Anne Lehmkuhl. It is a compiled tree and is not adopted. It is assessed in full at What Was Searched: it agrees with the allegation this archive already holds, it loses to a gravestone in three places, and it fails its own arithmetic in one. Two things in it are worth working. FIRST — JOHAN CASMIER KOLBE, listed ninth of the Settler's thirteen children. There is no Casmier, Casimir or Kasimir anywhere in this archive. If he existed he was baptised somewhere, and the Colesberg, Burgersdorp and Graaff-Reinet registers are ones this archive already reads. Test him before believing him: a compiler who splits one man into two, as this one may have done below, can equally invent one out of a misread hand. SECOND, AND BETTER — THE FORENAME “RECTOR”. This archive holds ANTON RECTOR KOLBE BARRY, born 3 November 1882 and baptised at Burgersdorp in 1883, and calls him *“THE KOLBE NAME IN A FOURTH GENERATION … the furthest down this archive has traced it”* — while saying nothing at all about the Rector, because there was nothing to say. The report lists PETER RECTOR KOLBE and BENJAMIN KOLBE as two children, where this archive holds one PETER BENJAMIN KOLBE carrying no dates and the single line *“Of the Kolbe children.”* A great-uncle called Rector would explain a great-nephew called Rector, and this archive's namesake reasoning is exactly how the Fortunatus name was traced across three families. But it cuts both ways: the report may have split one man in two, which would make “Peter Rector” and “Benjamin” a single Peter Benjamin and the Rector unexplained again. The paper that settles it is a baptism, and Peter Benjamin Kolbe is one of the five Settler children whose records this archive has never read. WHY THIS ROW IS NOT “CHASE A TREE”. Every other name in the report is one this archive already holds, which is what makes the two exceptions interesting rather than noise — a source that agrees with you everywhere you can check it has earned a look where you cannot. 21 SEPTEMBER, ATTEMPTED — AND EVERY ROUTE TO A REGISTER WAS SHUT. Recorded in full at What Was Searched. FamilySearch answered the search URL with its sign-in page, and this archive does not enter credentials. Geni served an empty body to both a fetch and a browser. eGGSA's gravestone library refused navigation — which mattered, because a stone naming a Rector in that generation would have answered as well as a baptism. Nothing is recorded against either name: no search of any register took place, and a question that could not be asked is not a question that came back empty. AND THE ARCHIVE NEARLY PUBLISHED A NEGATIVE IT HAD NOT EARNED. Three archive.org full-text queries for *Rector Kolbe*, *Casmier Kolbe* and *Peter Benjamin Kolbe* returned empty bodies. A control was run before any of it was believed: the two full-text endpoints answer HTTP 000 and the third answers 404, while archive.org's own search answers 200. The host is fine; its full-text API is gone. That is row 132's lesson arriving again the same day — treat a zero from an unverified object as no result, not as a negative. WHAT WOULD UNBLOCK THIS ROW: David signed in to FamilySearch, or row 93's affiliate library, which is already open for film 008004264 and could take both questions in the same visit. The five Settler children whose baptisms this archive has never read are the target, and Peter Benjamin Kolbe is one of them. 21 SEPTEMBER, ANSWERED — AND BOTH ANSWERS WERE ALREADY IN THIS BUILDING. After a day of trying to reach registers that were all shut — FamilySearch signed out, Geni empty, eGGSA refusing — both names were settled from this archive's own shelves in twenty minutes. (a) JOHAN CASIMIR EXISTED AND THIS ROW WAS WRONG ABOUT HIM. The claim above that there is no Casmier, Casimir or Kasimir anywhere in this archive is false and is corrected. He is named in BOTH his parents' death notices — the Settler's, signed M. Kolbe at Colesberg in January 1845, where he is the fifth of ten; and Margaret Downing's of 25 November 1867, signed by G. D. O. Werdmuller, which marks Mary Elizabeth *“(dead)”* and does not mark him, so he was alive at about twenty-three. Both have been transcribed here since 10 September, and he is on the Kolbe page with the variant “Casiner” already recorded. He was missing only from `people.js`, which is the one place the search was run. A search of the index is not a search of the holdings. He now has a record. AND THE NAME OUTLIVED HIM. Three men of it sit in this archive's own Register out of South Africa, Civil Marriage Records, 1801–1974 — born 1863, born 1934, and one undated. Not kin, not claimed — but a compiler does not invent a forename that three other men carry in the same country's marriage registers over ninety years. (b) ONE MAN, NOT TWO — AND THE RECTOR IS EXPLAINED. This archive carried him as “KEETOR Peter Benjamin”. Keetor is not a name; Rector is. The report's PETER RECTOR KOLBE is the same man with the forename read correctly, so on this one point the weaker source is the better witness — and the report then splits him into PETER RECTOR and BENJAMIN, inflating its own count, which is exactly the fault this row warned might be there. A great-uncle called Rector would explain ANTON RECTOR KOLBE BARRY, born 1882, whose Rector this archive has never accounted for. BUT NOTHING IS ADOPTED, AND THE DOCUMENTS ARE AGAINST HIM. He is absent from the 1845 notice's ten and from the 1867 notice's survivors, and a boy born in 1840 would have been four when his father died — a minor child's name is precisely what a death notice records. He died before 1845, or he does not belong to this family. Two compiled sources agreeing about a forename are still two compiled sources. WHAT IS LEFT IS ONE ERRAND: a Colesberg baptism for 1840, and a Bethulie or Colesberg baptism for Johan Casimir, neither of which has been read. RECTOR returns nothing in the surname harvest, and that is NOT recorded as a negative — the Kolbe harvest is 400 of 1,585 records, 25%, which is too little ground to report a nil from. AND THE 'WHAT WOULD UNBLOCK IT' ABOVE IS ITSELF PROBABLY WRONG — TRY A FRESH TAB BEFORE ASKING ANYBODY. This row says the remedy is David signed in, or row 93's affiliate library. That was concluded from a sign-in page seen in one browser context. The Lerena session has since shown the account was never signed out — another session got 3,652 and 3,967 results in the same minutes with the account name in the nav bar, because each browser context has its own cookie jar and one lost auth and recovered on its own. Their rule, and it is the right one: before asking a person to fix something, open a second tab. So the Casimir baptism and the Colesberg 1840 baptism may be reachable today without anybody signing into anything. Not retried here because Chrome became unreachable.
178. Next — ready to start David
Open one genealogy.com thread in an ordinary browser — it is titled for the Settler and this archive cannot reach it
A two-minute job for a person and an impossible one from here. `genealogy.com/forum/regional/countries/topics/southafrica/5263/` is a forum thread headed GEORGE AUGUSTUS KOLBE — the Settler by name. Threads like it are where descendants leave addresses, family bibles and photographs, which is the one category of source this archive has almost none of. It is behind bot protection. A direct fetch returns two bytes; an ordinary fetch returns HTTP 403; a browser lands on a Cloudflare interstitial. Working around bot detection is not something this archive will do, so the thread is recorded as unread rather than empty — a null from a challenge page is not a null. It opens normally for a human. What is wanted is simply: who posted, what they claim, and whether anybody left a way to reach them.
179. Next — ready to start the Blazevic and Luwinski sessions
Four characters the shared name-fold cannot see — and why this archive is not the one to fix it
RAISED BY THE BLAZEVIC SESSION ABOUT ITS OWN MATCHER, AND IT IS TRUE OF THE KIT. `kit/tools/namefold.py` folds by `unicodedata.normalize("NFD", …)` and then dropping combining marks. That works for every accent that decomposes and silently fails for the ones that do not. Tested: `Blažević → blazevic` ok · `Żubrinić → zubrinic` ok · `Défranceschi → defranceschi` ok — but `Đurić → đuric`, `Łukasz → łukasz`, `Søren → søren` and `Straße → straße` all come through unchanged, because đ, ł, ø and ß are single codepoints carrying no combining mark. On this estate that is Blazevic (Đ) and probably Luwinski (Ł); the rest of the archives are ASCII and unaffected. THIS ARCHIVE IS NOT FIXING IT, AND THE REASON IS THE ESTATE'S OWN RULE. *Folding is allowed to miss; it is not allowed to join* — written after a doubled-letter rule merged Roos and Ross, two different families. **Mapping `đ→d` makes the fold join *Đurić* and *Durić*, and whether those are one surname or two is a judgement about Croatian names belonging to the archive that holds them, not to a Booyzen session editing a shared tool. The mechanism is mine to report; the join is theirs to decide. The offer stands: name the characters and they go in the kit. THE OTHER HALF OF THAT SESSION'S FINDING DOES NOT APPLY HERE, AND IT IS WORTH SAYING WHY. Blazevic found it had been publishing on a surname test rather than a descent test, so a daughter's children — blood, but carrying their father's name — fell off the site; its root person's own first cousins were missing. Booyzen has no scope predicate at all: the `l` field is a label, not a filter. And the census of lines is the proof that descent is already followed through daughters — 39 Barrys, 72 Mountjoys, 15 Daschners, none of them the root surname, all of them here because a Kolbe daughter married out and the line was followed. A test this archive passes for a reason, rather than by luck, is still worth writing down.**
180. Done — finished and published this archive
The navigation in the excerpts is gone and the file converges — what is left is the narrative leak
FOUND BY FOLLOWING A ONE-LINE CHANGE IN `namefold.json` TO ITS SOURCE. A new fold entry appeared — `kolbê → kolbe` — which turned out to be a genuine FamilySearch transcription *(“Pieter Jacobus Kolbê”, South Africa Civil Marriage Records)* and not a typo. Asking where it had come from is what turned up the rest. THE READER-VISIBLE HALF. `dossiers.py`'s `page_text()` strips `<script>`, `<style>`, `<nav>` and `<footer>`. The skip link and the site header are in none of those, so they are harvested into excerpts and printed verbatim on person pages: *“…A. M. Kolbe of Piet Retief — The Booyzen Archive Skip to the content The Booyzen Archive Menu Start Home…”* 283 of 9,008 excerpts — 3.1% — carry it: “Skip to the content” in 240, “Menu Start Home” in 271, the title string in 245. A reader sees the navigation quoted back as though it were prose about their ancestor. THE HALF THAT REACHES THE NAME-FOLD. `namefold.py`'s `corpus()` reads every `*.json` in `src/data` bar a NARRATIVE deny-list, and `dossiers.json` is not on it. So the corpus is 391,506 tokens of which 231,968 — 59% — come out of the generated file, and the words that reach it *only* that way are led by People 1407, Timeline 478, Blood 386, Spine 377, Others 369, Places 346, Evidence 346, Start 344, Skip 344, Menu 344, Search 256 — page furniture, in a function whose docstring says *“every word the archive's own records actually contain”*. AND THE NARRATIVE GUARD LEAKS BY THE SAME ROUTE. It excludes `worklist.json`, `searched.json`, `corrections.json` and the rest by filename — but those pages are built, scraped and stored, so 187 excerpts (2.1%) come from `/changes/`, `/corrections/`, `/open-questions/`, `/method/`, `/letters/`, `/searched/` and `/worklist/`. The prose the deny-list exists to keep out walks back in through the generated file, which is precisely what its own comment warns of. DO NOT FIX IT BY EXCLUDING `dossiers.json` FROM THE CORPUS — MEASURED, AND IT COSTS MORE THAN IT SAVES. Only three words would be lost, but they are BRONKHORST, BOOÿSEN and BOOŸZEN, two of them ÿ-spellings of this archive's own surname, for which `key()` carries a `ÿ→y` rule on purpose. **`corpus()` reads only `*.json`, so prose written directly in an `.astro` document page reaches the fold by no other path. The generated file is currently the only bridge from page prose to the fold; cutting it blinds the fold to every spelling that lives in a document. THE ONE EDIT THAT FIXES BOTH HALVES is stripping the header and skip link in `page_text()`, alongside the excerpt-section work already in flight. 21 SEPTEMBER, RESOLVED ON THE CHROME HALF, AND THE FILE NOW HAS A FIXED POINT. Generate → rebuild → generate, five cycles: 24,981 → 14,115 → 13,444 → 13,455 → 9,209 → 9,554 → 9,554 → 9,554, with changed = 0 of 347 on cycles three, four and five. It does not move any more. Excerpts quoting a count fell 1,732 → 12. AND THE PUBLISHED NUMBER WAS WORSE THAN I SAID.** I estimated the inflation at *“close to double”*. It was 24,981 against a true 9,554 — two and a half times. Every person page has carried a mention count roughly 2.6× the real figure, and *which* quotations appeared was arbitrary. The cause is registered in the corrections list under `circular`; the figure is not. A reader is owed the number, not only the reason. THERE WAS A FOURTH CARRIER OF THE CHROME AND I MISSED IT. I named the skip link and the site header. `page_text()` also read the `<title>` out of the page and never removed the element, so the title survived into the body text as well and every affected excerpt carried it twice over. `<head>`, `<a class="skiplink">` and `<header class="site">` now all go, one variant of each across 796 pages. In the converged file the chrome was 344 of 5,864 excerpts — 5.9%, proportionally worse than the 3.1% measured on the committed one. That page's text now opens “Kolbe · A.M. Kolbe of Piet Retief · On the same stone, and known by two initials only.” THE HEADER STRIP ALSO FIXES THE CORPUS AT THE RIGHT LAYER, which is better than the exclusion I nearly recommended: People, Timeline, Blood, Spine, Others, Places, Evidence, Start, Skip, Menu and Search only ever reached the fold as chrome, so removing them at source leaves the bridge from page prose to `namefold` intact. The fold was not wrong to read the file; the file was wrong to contain furniture. WHAT IS STILL OPEN, AND IT IS THE WHOLE OF THIS ROW NOW: THE NARRATIVE LEAK. 187 excerpts harvested from `/changes/`, `/corrections/`, `/open-questions/`, `/method/`, `/letters/`, `/searched/` and `/worklist/` — prose *about* the archive re-entering as prose *of* the archive, which is exactly what the deny-list in `namefold.py` exists to prevent and cannot see, because it works on filenames and these arrive as scraped pages. It needs a decision rather than a patch: whether those routes belong in the scrape at all. A person genuinely is discussed on `/changes/`, so excluding them costs real mentions; keeping them feeds narrative into the fold. Left stated rather than half-done. 21 SEPTEMBER, COSTED — THE DECISION IS NOW A CHOICE BETWEEN TWO KNOWN PRICES RATHER THAN AN OPEN QUESTION. Measured on the converged file: 314 of 5,863 excerpts (5.4%) come from narrative routes. 127 of 347 people are affected. NOBODY DROPS TO ZERO. Not one person's page would turn into *“not yet discussed on any page”*. The largest single loss is james-mountjoy, 8 of 23 (35%), then johan-gottlieb-kolbe 8 of 26, magdelena-elisabeth-de-beer 8 of 26. AND THE COSTING SHOWS THE QUESTION WAS SLIGHTLY WRONG. A dossier and a name-fold want opposite answers out of the same file. For a reader, a person discussed on `/changes/` genuinely is discussed there, and the excerpt is honest. For the fold, that is the archive writing *about* a name, which is exactly the standing the NARRATIVE deny-list exists to refuse. One file is serving two purposes with one rule. RECOMMENDATION — EXCLUDE THE NARRATIVE ROUTES FROM THE SCRAPE. Nobody is orphaned; a mention on `/changes/` is the archive talking about itself rather than evidence about a person; it closes the fold leak at source with no new mechanism, in the same place `/who/` and `/places/` are already excluded; and 5.4% is a small price for the least informative excerpts on any dossier page. The one worth arguing over is `/corrections/` — that a person's record was withdrawn or amended is something a reader should meet, but it belongs as a link to the corrections register rather than as a quotation harvested out of it. The alternative, if those mentions are wanted, is to make the fold structure-aware — `dossiers.json` records a `page` per hit, so `corpus()` could parse rather than regex the file and skip narrative-sourced hits. That keeps every reader-facing excerpt and still closes the leak, at the cost of the one tool in the kit that currently reads its inputs as flat text. 21 SEPTEMBER — DONE, AND THE MEASUREMENT IS WHY IT WAS SAFE TO DO. Nine routes are now skipped in `tools/dossiers.py` where `/who/` and `/places/` already were: /changes/, /corrections/, /open-questions/, /method/, /letters/, /searched/, /worklist/, /research-log/ and /errands/. 330 of 5,864 excerpts, 5.6%, across 128 of 347 people — and nobody is orphaned, not one page becoming “not yet discussed on any page”, which is the failure this tool has already caused once. Largest single loss is james-mountjoy, 9 of 23. Convergence re-tested rather than assumed: five cycles at 9,141 mentions, changed=0 of 347 after the first, `dossier-drift` zero, every gate green. The chrome words are gone from the fold at source — People, Timeline, Bloodline, Start, Skip, Menu, Search reached it only as furniture — while BRONKHORST, BOOÿSEN and BOOŸZEN are untouched, the three that reach the fold through this file and no other way. `/corrections/` went with the rest and it is the one worth arguing over: that a record was withdrawn is something a reader should meet, but as a link the page carries deliberately, not as a quotation a scraper lifted out of the register without knowing what it was reading. `/disputed/` and `/name/` were deliberately kept — they read like registers and are not. The alternative was declined on cost: teaching the shared fold's `corpus()` to parse a schema instead of reading flat text is a large bill in a tool eight archives share with eight different data shapes. Commit `db1deeb`. The measurement, the reframing and the recommendation are the landing-site session's.
181. Done — finished and published this archive
I broke the build twice in one commit, and one of the two was the gate's fault as much as mine
BOTH FAULTS WERE MINE, BOTH WERE IN `826bba1`, AND THE OTHER SESSION SPENT THREE BUILDS FINDING THEM BY HAND. Written down because this archive records its own mistakes. THE FIRST WAS A VERDICT THAT DOES NOT EXIST. The searched register allows five — yes, part, no, void, blocked — and `searched.astro` indexes `badge[k][1]`. I wrote `"NOT READ"` into the badge slot of the genealogy.com row, which is `undefined[1]`, and the whole build dies with /searched/ naming none of its 687 rows. `blocked` already existed for precisely that case and is what the row now carries. The worse half is how it shipped: the isolated build I ran was taken *before* that row existed, and afterwards I ran only `check:worklist` and a JSON-validity check — neither of which touches this schema — and then reported the work as verified. A build that did not cover the change is not a build. An unknown verdict is now gated and names the set and the row index. THE SECOND WAS PROSE THAT QUOTED MARKUP. The graves.co.za row reproduced a parked domain's redirect script verbatim, so `href="/lander"` sat in the page text and the link checker read it as a link to a page nobody had claimed existed. AND THAT SECOND ONE IS NOT PECULIAR TO ME, WHICH IS WHY IT BECAME A KIT CHANGE. `checkarchive` matched `href="` anywhere in the body. Lerena carries 207 of the same shape on `/register/` alone — escaped tags inside transcript notes — an `a` element carrying class `ark` and an external address, written as an escaped tag rather than a real one, so it is prose that looks exactly like markup. Those are external so the checker skips them and nothing has failed there; an internal path in that position fails a build over a sentence. The href must now sit inside a real tag, at both extraction sites — the link check and the reachability crawl, which would otherwise invent graph edges out of prose. MEASURED BEFORE TAKING IT, because a link checker that goes quiet is worse than one that cries wolf. Across five built archives — Falco 704,991 internal hrefs, Blazevic 406,699, D'Arcy 156,811, Mazza 116,232, Lerena 68,209 — 1.45 million — the tagged form misses none of them and drops only the quoted ones. Re-run after: Falco, Lerena, Blazevic and Mazza all green, and Lerena's full output including warnings is byte-identical before and after, so no page lost its reachability either. Kit `b86012d`, repinned here. 21 SEPTEMBER, A CORRECTION TO THIS ROW THAT BELONGS TO THE OTHER SESSION AND NOT TO ITS AUTHOR. This row first said `4027ccc`, and it said so because I told them it was `4027ccc` — twice, in writing, and again in a commit message — having read the hash off `git log -- searched.json` and taken the top entry without testing it. The commit is `826bba1`, *“Two gravestone databases we had never opened”*. Two independent proofs. `git log -S'"NOT READ"' -- site/src/data/searched.json` names 826bba1 and not 4027ccc; and GitHub Actions is decisive on timing — 4027ccc passed at 21:04:33 and 826bba1 is the first failure, at 21:09:20, followed by four more until the fix. 4027ccc could not have broken a build it went green on. THIS IS THE THIRD UNTESTED REFERENCE FROM THIS ARCHIVE IN ONE DAY — row 174's `6e4a33f`, the archive.org identifiers in row 132, and now this — and it is the worst of the three, because the other two were mine and stayed mine, while this one was published in somebody else's confession of their own fault. A wrong hash in a row about being careful is the whole lesson. `git log -S` on the actual string is what settles which commit introduced a value, and the CI timeline settles it a second time for free. AND THE HALF OF THAT CORRECTION THAT BELONGS BACK TO THIS SESSION: I CHECKED THE HASH MYSELF, GOT THE WRONG ANSWER, AND REPEATED THEIRS AS THOUGH I HAD CONFIRMED IT. The check I ran was `git show <commit>:searched.json | grep -c '"NOT READ"'`. It returned 1 for `4027ccc` and 2 for `826bba1`, and I read the 1 as my defect. It was not. `NOT READ` is a legitimate value in slot [1], the verdict *headline*, and the Volksrust camp-register row has carried one for days. I counted a string that occurs lawfully in a different column and read the count as identifying my row — so I had independent evidence in front of me and it agreed with the wrong answer. The check that actually settles it reads the position, not the string: parse the JSON and count rows whose slot [3] is outside the five the badge map allows. `4027ccc` has zero. `826bba1` has exactly one — row 6 of the Barry register set, which is the genealogy.com row I appended. SO THE DAY'S PATTERN CLAIMED FOUR, NOT THREE, AND THE FOURTH IS THE ONE I FOUND MOST CONVINCING AT THE TIME. A grep with a plausible count, believed because it was a measurement. A measurement of the wrong column is not a measurement of the column you meant, and the commit message at `9d15d21` still says both faults were in `4027ccc`. That sentence is wrong, it is pushed, and it is not being rewritten — the record corrects it here, which is what this list is for. 21 SEPTEMBER, A SMALL EDIT TO THIS ROW BY THE BOOYZEN SESSION, AND IT IS THE ROW'S OWN SUBJECT BITING IT. This row reproduced the offending markup verbatim to explain it. Two gates here then refused the build over it: `links.py` read the quoted `href="` as a link — the identical fault this row is about, in this archive's local copy, which the kit change never reached — and `drift.py` refused because the escaped less-than was escaped a second time and reached the reader as literal text — and this sentence will not name the entity either, because writing it here reproduced the fault a third time and refused the build again. The passage now describes the markup instead of reproducing it, which is what the graves.co.za row was changed to do earlier the same day for the same reason. Nothing about the finding is altered. The lesson is the row's own: a sentence that quotes markup is markup as far as anything scanning the page is concerned, and the only safe way to write about an `href` is not to write one.
182. Done — finished and published this archive
A pin can sit in the right country and still be a lie about precision — the D'Arcy session's detector, run here
FOUND BY THE D'ARCY SESSION ON THEIR OWN ATLAS, AND IT IS THE BEST KIND OF FINDING — one this archive's existing gate could never have seen. Twenty-one different places shared the single point 52.53102, -1.26491, carrying 34 people, and every one was stamped exact. That point is the geocode for the word “England”. On it sat Berkeley — the place that archive is about — beside London, Portsmouth, Somerset, Clarkenwell, Theobalds Palace and sixteen more. It repeats down the administrative levels: five on the Queensland centroid, four on Gloucestershire (including Wotton-under-Edge, where their Lydia was born), three on Staffordshire. In all 36 inherited pins carrying 43 people, 30 claiming to be exact. WHY THE KIT'S GATE CANNOT SEE IT, AND IS NOT WRONG NOT TO. `checkplaces.py` refuses a pin outside the country its own text names — it is a wrong-COUNTRY test, written after the Defranceschi archive put Pag in the Philippines, 9,000 km out. An inherited pin is in the RIGHT country at the wrong precision, so it passes that gate cleanly. Two orthogonal faults, and an archive needs both checks. THE DIAGNOSIS IS NOT DAMAGED STRINGS, WHICH IS WHAT MAKES THE DETECTOR PORTABLE. Their first thought was broken export strings, but “Teddington, Middlesex, England” and “Wotton-under-Edge, Gloucestershire, England” are perfectly well formed and land on the country anyway — and Solihull lands there spelled correctly and spelled Solihul. The gazetteer has not heard of them and answers with the smallest thing it did recognise, and the confidence field records that the LOOKUP succeeded, not that it answered the question. A string-tidying rule would have missed all of it. RUN HERE, AND THIS ARCHIVE IS CLEAN. 60 of 65 places carry coordinates; two points carry more than one place and ZERO pins are inherited. The two shared points are honest: Manorowen, a farm, sits on Barkly East, and Wodehouse, a district, sits on Dordrecht — and in each case the borrower is flagged `approx`, which is a pin saying “about here”, which is true. SO THE FAULT IS NOT SHARING A COORDINATE. IT IS SHARING ONE WHILE CLAIMING TO BE EXACT, and that is what is now gated. `checkregisters.py` check 18 refuses a build where two places sit on one point and neither says `approx`, and separately where a place's coordinate equals that of another whose whole name is a comma-tail of its own — the D'Arcy session's own test, decided on the recorded string and never on the coordinate alone, so it cannot accuse two towns that genuinely stand close together. Verified both ways: dropping Manorowen's `approx` names Barkly East and Manorowen and the point; adding a “Somewhere, Barkly East” on Barkly East's coordinate names it as inherited; restoring is silent. KEPT LOCAL AND NOT PUT IN THE KIT, deliberately. Eight archives share that tool and only two have measured this; the case for promoting it belongs to whoever measures it across all of them, which is how the `href=“` fix was decided the same day. The Defranceski session has already ported the detector in about twenty lines and found four Croatian villages on their district towns — and there the cause was entirely different, an upstream FamilySearch KML rather than any geocoder of theirs. The detector found the symptom without needing to know the cause, which is the whole argument for it.
183. Running — a machine or a person is at it now this archive
Ninety-six of 329 sources in the estate said where they were; 182 do now
The first version of this row was wrong and the error is the point of it. It said 58 of 208 sources carried an outbound link and that all 58 were Defranceski's. The scan only looked for `sources.json`. Lerena keeps 121 sources in `sources-consulted.json`, 38 of them already linked, under a page heading reading *Every collection searched, with its link*; Luwinski wrote the host after almost every title in prose. The gate reported *no sources.json — nothing to check* and PASSED, so it produced a confident number that was false, and it was quoted in four commit messages. It now fails loudly when it cannot find its subject, proven by injection both ways.
True figures: 96 of 329 before, 182 after, plus 13 that say where the paper is. D'Arcy 0→18 of 18 · Falco 0→22 of 26 · Luwinski 0→17 of 24 · Defranceski 58→66 of 85 · Mazza 0→12 of 25 · Blazevic 0→9 of 12 · Lerena 38 of 121 unchanged, already done · Booyzen 0 of 42, and see below.
This archive is last because its `sources.json` is not a bibliography — it is four tables of HOLDINGS. Its real source list is prose in six other sections, five of which now link. Turning the rest into rows is a restructuring and wants David.
184. Running — a machine or a person is at it now this archive
Nobody updates the sources page while they research, and no gate could tell
Asked whether each archive keeps its sources page current. It does not. Commits to each archive's data since its sources page was last touched: Lerena 302, Defranceski 121, Luwinski 93, Blazevic 64, Booyzen 57, D'Arcy 29. Two attempts to turn that into a gate were wrong before the third was right, and both are recorded in `checksources.py` rather than deleted. Matching free text for source-shaped strings offered *“Argentina-South Africa relations.”* as a missing source. Matching identifiers by shape — arks, film numbers, KAB and NAA references — offered 7,502 of them for Booyzen, every one a real identifier and not one of them a finding, because a sources page names collections and nobody should list 7,502 films. Both failed the way this estate already has written down: *a check that tests something else and reports it as this.* What works is hosts. Every domain an archive links to in its data, against the domains its sources page names. Undeclared today: D'Arcy 58, Defranceski 74, Blazevic 17, Lerena 11, Falco 3 — including the Antenati host Falco has a whole page about — Booyzen 3, Mazza 3, Luwinski 0. Tens rather than thousands, each fixable by adding one line, and it is the same line that makes the source reachable.
Three faults the gate found, two of them mine. Mazza's `sources.json` is generated from a TSV and silently reverted twelve verified urls on the next build — the page would have looked identical. Lerena was invisible to the check entirely. And `components.py` has been counting seven archives in an estate of eight because Luwinski was never in its list, so every figure it printed was short by one.
185. Next — ready to start this archive
Defranceski draws its family lines as journeys; six archives have the same thing as a list
Raised from the Defranceski families / lines drawing: one band per line, each a row of dated waypoints running left to right, fanning at the end into where that line went, with a confidence stamp under the title (`DOCUMENTED`, `DOCUMENTED FROM 1667 · ORIGIN NOT FOUND`, `DOCUMENTED · NOT THIS FAMILY`) and, underneath, two honest footers — households of the name not yet placed on any line, and stock attached to no line here at all. Nothing else in the estate draws this. `Trees` states each line in the form its evidence supports but does not move it through places; `BranchRivers` is Defranceski's alone; the rest hold their lines as prose or a table. Compare every archive's families/lines page against it and plan the harmonisation — the parts to test are whether a line is a sequence of PLACES (Defranceski) or of PEOPLE (most others), and whether the fan-out at the end is a claim the archive can carry.
186. Next — ready to start this archive
Of 447 pages in the estate, 39 say which sources they rest on — and all 39 are Defranceski's
Found while measuring the link gap, and it is the bigger finding. Defranceski does not only have the best sources page; 39 of its pages carry a block declaring which sources they rest on, by id, and the id must resolve against sources.json or nothing renders. Blazevic 0 of 65, D'Arcy 0 of 79, Falco 0 of 58, Mazza 0 of 45, Lerena 0 of 43, Booyzen 0 of 41, Luwinski 0 of 29. This is *why* Defranceski's sources page has not drifted while the others have — a page that names its sources by id cannot drift, because the id has to exist. Promoted to the kit as `SourceBlock`, which drops an unresolved id and counts it onto `data-missing` so a gate can fail a build over a page claiming a source the archive does not hold. Booyzen's own 41 pages declare nothing. That is this archive's work, not a note about someone else's.
187. Done — finished and published David
Send Anne Brest the summary she asked for — what the work is, why the files were wrong, and what her invoice bought
SHE ASKED FOR IT AND DAVID PROMISED IT, on the thread, the morning the payment went: *“I will send a summary when I'm home.”* Her own words were *“Maybe you can tell me what is is you are trying to do and why the files were the wrong files. I can help with info if you need it.”* She is offline until Monday night or Tuesday, so it will be waiting rather than late. DRAFTED AT `requests/anne-brest-summary-what-and-why.md`, WITH THE INVOICE AS THE FOCUS. A first draft gave her work a single paragraph and then spent its length on this archive's own mistake; that was the wrong shape for a letter to somebody who has just halved her bill. The draft itemises what the R5,900 bought, says which parts survive the cancellation, and says plainly that the verification is work that would have had to be paid for whenever the visit happened. IT CORRECTS A NUMBER SHE WAS GIVEN — two of six became three, plus a fourth that is not a file at all. IT WITHHOLDS THE LUWINSKI DNA FINDING deliberately: an inference about a living man's parentage is not a contractor's business, and the archive names him and nothing else. IT MAKES ONE ASK, the paper MHG index for the 1920s, because she offered and because it is the single question nobody outside that building can answer. 21 SEPTEMBER, REDRAFTED TWICE MORE AND IT NOW ASKS FOR NOTHING. The second draft itemised the invoice properly but ran long; the third was short and ended by commissioning the paper-index lookup and offering to pay for it. David is not committing to paying for more work, so the fee offer and the promise of a corrected list are both gone. What is left states the goal, marks her seven lines still need / don't need, names the two gaps — MHG 40293 and the correct 1920s reference behind 52653 — and stops. She offered information, not labour, and the letter takes her at that word. SENT 21 SEPTEMBER 2026, Gmail message id `1a0c3a45abd88f27`, as a reply on her own thread. She is offline until Monday night or Tuesday, so it is waiting rather than late. Nothing is owed by either side: the invoice is paid, no new work is commissioned, and the two gaps are stated as gaps. Whatever comes back is hers to volunteer.
188. Done — finished and published David
Anne Brest has the links — the children's site, Booyzen, Luwinski and the Record Atlas
SENT 21 September 2026, Gmail message id `1a0c3b1cdc859b23`, an hour after the summary and on the same thread. A postscript that asks for nothing and says so. All four URLs tested HTTP 200 before sending — this archive's rule about untested references applies to its own links too. Luwinski was included on David's explicit instruction, after the consequence was put to him: the summary had withheld the DNA finding as not a contractor's business, and that archive publishes `/dna/` itself, so the link reaches it. A choice made rather than an oversight, and recorded as one. Nothing is now outstanding with Anne: invoice paid, visit stood down, both questions answered, no work commissioned.
189. Running — a machine or a person is at it now this archive
Sixty-seven pages in the estate drew a shared idea by hand; there is a gate on it now
`checkshared` has watched duplicated components since 14 September and could never have found the worst duplication in the estate, because the thing duplicated was a page: `places/[slug].astro` was 85 lines with the same md5 in Falco, Blazevic, Mazza and D'Arcy — one file copied four times, comment and all.
`checkpages.py` measures it: 37 page names carried by three or more archives, 67 archive-pages drawn by hand on the day it was written. It discovers archives by globbing `*/site/src/pages` rather than keeping a list, because `components.py` kept one, spelled D'arcy with the wrong case and omitted Luwinski entirely — so every count it ever printed was short by one archive.
It is a ratchet, not a wall. `--max-bespoke` fails only when the number goes UP. What it catches is the cheap moment: adding a new hand-drawn page for an idea three archives already share fails the build that adds it. Wired into all eight — this archive at 4.
And an archive can refuse, in writing. `.harmonise-exempt` takes a page and a reason and the reason prints on every run. Reading the candidates is what the threshold cannot do: `namesakes` is carried by three archives and is three different pages — this archive's is about FORENAMES, Defranceski's collects notable strangers of a rare surname, Falco's argues for not having one. `research-log` is four archives and at least three things. A shared page name is not a shared page.
Done so far: 67 → 50 bespoke, 6 refused in writing. `PlacePage`, `Errands` and `Shotlist` promoted. Left, largest first: `dna` 6 archives 3,681 lines · `name` 7 archives 1,846 · `people/[slug]` 7 archives 1,676.
190. Done — finished and published the shared kit
The kit's sitemap.py pins site/dist, so a session cannot fully isolate its build
RAISED RATHER THAN PATCHED, BECAUSE EIGHT ARCHIVES SHARE THE FILE. `kit/tools/sitemap.py` computes `dist = os.path.join(site, "dist")` from the working directory and takes no argument, so it writes `sitemap.xml` into `site/dist` no matter where the pages actually went. WHY THAT MATTERS NOW: this archive can give a session its own output directory with `ARCHIVE_OUT`, which stops a concurrent build in another session from emptying the one its gates are reading. Everything honours it — astro, `check:kit`, `check:living`, `check:evidence`, `gallery.py`, `dossiers.py`, `links.py`, `checkatlas.py`, `checkregisters.py` — except the sitemap, so an isolated run trips `check:kit` with *“robots.txt points at sitemap.xml and no sitemap.xml was built”*. The gate is right; the tool is the problem. THE FIX IS ONE ARGUMENT — `--dist`, defaulting to `dist`, the same shape every other kit tool already takes. Not made here, on the precedent set on 21 September: a kit change gets measured across the archives that share it before it is promoted, and this one has been measured in exactly one. Until then an isolated build is for reading gates, not for deploying, and `build.sh` says so when it isolates. THE DEFAULT IS UNCHANGED AND MUST STAY SO: `.github/workflows` uploads `path: site/dist`, so a different default would deploy an empty site. 22 SEPTEMBER — RAISED WITH THE SESSION THAT OWNS THE KIT (the landing site), with the measurement done first because that is the precedent it set. Seven of the ten projects on this machine use `kit/tools/sitemap.py` — Blazevic, Booyzen, D'Arcy, Defranceski, Falco, Lerena, Mazza; Luwinski, Our Family and Record Atlas do not. And the case was put honestly rather than at its strongest: this is not breaking anyone today. Booyzen is the only one of the ten that sets `outDir` at all, so the other six have never met it. It is a capability none of the seven has, not a fault in the field — a weaker case than the `href="` fix, which carried 1.45 million hrefs of live exposure, and saying so is part of asking. The change is inert for the other six: an optional `--dist` defaulting to `dist` leaves them byte-identical because none of them passes it, and `checkarchive.py`, `checkliving.py` and `checkevidence.py` already take exactly that flag — sitemap.py is the odd one out. Their call whether it earns a release. This row closes when it lands and is struck if they decide it does not. DONE THE SAME HOUR IT WAS RAISED — THE LANDING-SITE SESSION SHIPPED IT. Kit 40bf5a3: `sitemap.py` now takes `--dist`, defaulting to `dist`, and resolves an absolute path too, which is more than was asked for. The six other archives are untouched because none of them passes the flag, exactly as the request predicted. This archive's `sitemap` script now passes `--dist "${ARCHIVE_OUT:-dist}"` like every other `--dist` caller here. ISOLATION IS NOW COMPLETE AND TESTED END TO END. `ARCHIVE_OUT=dist-verify sh build.sh` → exit 0, 0 failures, 0 warnings, with `index.html` and `sitemap.xml` both in `site/dist-verify`. And the default path is unchanged where it must be: a plain `sh build.sh` → exit 0, with both files in `site/dist`, which is what `.github/workflows` uploads. A NOTE ON HOW THE KIT UPDATE ARRIVED, because it cost a confusing build. The repin landed while a build here was running — `check:pin` reported `9ff6320` when the measurement was taken and `40bf5a3` minutes later — and that run failed with an empty `dist` and no gate line, which read like a fault and was not one. A shared dependency changing under a running build is the same class of thing as a shared `dist` being emptied under a running gate, and it is the second instance today.
191. Done — finished and published this archive
package.json and package-lock.json are one change — staging half of a kit repin broke CI
COMMIT 88c90bc WENT RED AND THE LOCAL BUILD WAS GREEN THROUGHOUT. `check:pin` in CI: *“1 build script(s) call a kit file that is not in the installed kit — CI installs from the pin and will not find them either.”* THE CAUSE: `package.json` pinned the kit at 40bf5a3; `package-lock.json` still said 9ff6320. CI runs `npm ci || npm install`, and npm ci installs from the LOCKFILE, so it took the old kit and the scripts then asked it for a file it does not have. Fixed with `npm install --package-lock-only` — a two-line diff, nothing else moved — and green at `b0c9054`. AND THE PIN WAS NOT MINE TO CARRY. Another session repinned the kit while this one worked; I staged `site/package.json` for my own sitemap change and swept their pin in without the lockfile that belongs with it. Same fault as the swept commit of 21 September, at a smaller stage: staging a shared file whose diff I had not read. MEASURED ACROSS THE ESTATE, AND THE TRAP IS LIVE ELSEWHERE. Every archive's HEAD is internally consistent, so nobody's CI is broken — but seven archives are carrying that repin as TWO UNCOMMITTED FILES (Blazevic, D'Arcy, Defranceski, Falco, Lerena, Luwinski, Mazza). The sweep updated both; the trap is purely at the commit boundary, and the first session in each that stages `package.json` alone ships the same mismatch. Passed to the landing session. WHY THE LOCAL GATE COULD NOT SEE IT: `check:pin` compares the scripts against the installed kit, and this machine had the new one. It was right about this machine and silent about CI. The lockfile is the thing that encodes the difference, which is the whole reason it is committed. AND THREE WRONG INSTRUMENTS ON THE WAY — `git ls-remote` lists refs and not arbitrary commits, so its silence meant nothing; a `git fetch --depth=1 <sha>` probe called both the pin and the kit's real HEAD unfetchable, because GitHub refuses arbitrary-SHA fetches by default and a test that cannot distinguish its two cases has not been run; only a clone settled it. See the method.
192. Done — finished and published this archive
The spine must not link its names to dossiers — two of its six men share a slug
WRITTEN DOWN SO THE NEXT IMPROVEMENT TO THIS PAGE DOES NOT MAKE IT. The shared `Spine` component accepts an `href` per row, and linking every name on the spine to its dossier looks like the obvious first win. It is unsafe here. Generations five and six are both Petrus Jacobus Booysen, father and son, and they slug identically — they are one of only two shared-name dossiers in this archive, already merged and flagged by `dossiers.py`. Filling `href` mechanically would point two different men at one page and quietly assert they are the same, on the page whose entire purpose is to show a descent. AND IT IS NOT PECULIAR TO THIS ARCHIVE. The landing session measured it across the estate after this was raised: five of seven spines carry a repeated name. Falco has two Raffaele Falco, grandfather and grandson; D'Arcy repeats George Pitt D'Arcy; Mazza two; Defranceski three. A spine follows one line of men, which is precisely where reused forenames concentrate — so the slot that looks most obviously fillable is the one most likely to merge people. It is now in the kit's own trap notes. THE OTHER SLOTS, MEASURED RATHER THAN COUNTED. `bornPlace`/`diedPlace` render as `born <date>, <place>`, which is the string already in `born` — pure churn here, though the same slot is a genuine gain in Falco, where those are separate fields and absent from the page. `confidence` would print `doc` against all five and sit beside generation seven's dispute note reading as reassurance. `household` would duplicate the eight-child table this page already prints — an empty slot sometimes means the archive drew it better elsewhere. Only `trade` is additive, and only for two rows. AND A GATE WAS DROPPED ON THIS EVIDENCE. The landing session had proposed having `Spine` report how many optional slots an archive fills and ratchet it. A filled-slot count is a prompt to look and never a target to hit — ratcheting it would have pushed six archives toward the one unsafe slot to make a number go up.
193. Blocked — waiting on a person or an institution David
The reference for TAB MHG 40293 is incomplete, and the researcher says she knows the missing part
SHE VOLUNTEERED IT AND IT IS THE MOST VALUABLE SENTENCE IN HER REPLY. Of MHG 40293 — KOLBE, Frederick Fortunatus Downing, 1918, the one page this archive most wants — Anne Brest wrote: *“david, this file is available. . you have left out part of the reference. . I know what it is.”* This archive does not know what is missing and is not going to guess: the reference as held here is `TAB MHG 40293` and she is saying that is short of something. ASK HER WHAT IT IS. It costs a sentence and it is the difference between an order that resolves and one that does not — which this project has now got wrong twice in a fortnight, both times on a reference somebody else could see and it could not. 22 SEPTEMBER — NOT ASKED, AND DELIBERATELY SO. The Pretoria work is held off on cost, and the letter saying so asks her for nothing. Requesting free expertise in the same message that declines to pay is not something this archive does, so the missing part of the reference keeps until there is an order. It is the first thing to ask when there is one.
194. Blocked — waiting on a person or an institution David
The Pretoria estate files are held off on cost — the list is settled and nothing is on order
HELD 22 SEPTEMBER 2026, ON COST, AND SAID PLAINLY TO THE RESEARCHER. Anne Brest had marked up the list and offered to revise her invoice and proceed. Another order of that size is more than can be taken on now, so the work is stopped rather than drifted, and she has been released from holding a date, the files, or another favour at Preservation. THE LIST ITSELF IS SETTLED AND IS NOT THE BLOCKER — five items, four paid and one free: MHG 1095/48 (Booyzen, George Downing Mountjoy), MHG 5155/51 (Booyzen, Gert Jacobus), MHG 52653 (Booyzen, Jacobus Nichols, 1923 — a file, on her correction), MHG 40293 (Kolbe, Frederick Fortunatus Downing, 1918) and MHG 10245/69 (Wear, the free one). Every one survived a check against the free trees, which is what cut the original list by more than half. WHAT IS OWED TO NOBODY AND WHAT IS: Invoice 464 is paid, the visit is stood down, and no work is on order. The R5,900 stands as a credit she offered against a future list and was not raised in the letter — that is David's to take up if and when he goes back. WHEN IT RESUMES, ASK FOR THE MHG 40293 REFERENCE FIRST (row 193): she says this archive's version is incomplete and that she knows the missing part.
195. Done — finished and published this archive
The archive was quoting two documents at length and holding neither — now held, and not published
NOTHING HAD ENTERED `sources/` SINCE 16 SEPTEMBER, and nothing had entered the published `site/public/docs/` since the 13th — 200 images, unchanged for ten days, while the week's reading was quoted across `/documents/`, `/kolbe/`, `/sources/` and the timeline. THE TWO MOST-QUOTED DOCUMENTS OF THE WEEK EXISTED ONLY IN A DOWNLOADS FOLDER. The letter of 23 August 1830 (NLSA, MSB 751) and Schoeman's February 1992 DSAB typescript were read at full resolution, transcribed, tested line by line and cited on four pages — by an archive that held no copy of either. If those files had been tidied away, the evidence behind a week of published work would have gone with them. For an archive whose whole claim is that everything traces to a record, that is the gap that matters most. BOTH ARE NOW IN `sources/archive-scans/`, named to the house convention, tracked in git, and deliberately NOT published. `sources/` is a custody store that no built page references; `site/public/docs/` is the published one. Library-supplied material stays in custody rather than going onto a public website, which keeps the evidence without republishing what the NLSA sent. STILL NOT HELD, and each is somebody else's to supply: the Bell watercolour (MuseumAfrica / Edinburgh CSWC 60), the two Kolbe–McCabe marriage images, and Sibree's register page. A false positive worth recording: a search for “Bell” in the scans matched **Isa*bell*a** and nearly had this archive reporting it held a picture it has never seen.
196. Done — finished and published this archive
Row 167 was wrong, and the sentence it fixed was still false: this archive does hold a living person
* OVERTURNS ROW 167. * On 21 September this archive counted its living people, found none, and rewrote /register to say so from the count rather than from the rule. The count was honest and the conclusion was false. Tersia Booyzen is living, and this archive says so itself — she is recorded b = "living", the word sitting in the field a birth date would occupy.
* THREE READERS OF "IS THIS PERSON ALIVE" EACH MISSED HER, and each agreed with the others for the same wrong reason. * The kit's checkliving read living FLAGS, reported 0, and then passed its own self-test — the harvest worked perfectly on an encoding it never read. The kit's chart helper rendered «b. living» as though it were her birth date. And /register published, in bold, «No living person is in this register, and none is anywhere in this archive: of the 354 people it holds, not one is living», from a filter testing `living === true || conf === "living"`.
So row 167's lesson was right and incomplete. It said the fault was numbers typed instead of counted. The fault under that is that a number counted by a reader which cannot see the case is worth no more than a number typed — and two readers agreeing is evidence only when they CAN disagree.
* FIXED BY DECLARING THE VOCABULARY ONCE. * kit/data/living-words.json now holds how this estate writes down that somebody is alive — the flags, the words that mean living where a date belongs, the fields they can appear in, the 80-year rule — and both the gate and /register read it rather than each restating a subset. /register now publishes «One living person is in this register — Tersia Booyzen — of the 354 it holds» and that the rule is applied to her: her name is all this archive publishes. Naming her IS the rule, and she is named because she is a SOURCE — the Anglo-Boer War account and the name-change are her testimony, as told to her by her father.
Found by the D'Arcy session's query, which was to look for counts typed into sentences rather than computed into them. 228 such counts across eight archives; nearly all are readings of other people's holdings — films, museum registers, other people's trees — correctly typed and not computable. This was the one that was about us. Her name is also now a declared decision in decisions.json, so anyone appearing in or disappearing from that list stops the build.
197. Done — finished and published David
FamilySearch is signed out in this session's browser, and only David can sign it in
The browser pane restarted and lost the session. The identical search URL that returned 3,499 results this morning returns 0 now, captioned *No Results Found* — the fabricated null this archive has been warning about since 21 September, reproduced here under control. A fresh tab does not fix it: the session belongs to the pane. No session here will enter credentials on David's behalf, so this needs him for thirty seconds. WHAT IT HOLDS UP: the record arks for five of the Commandant's twelve children, who cannot be entered until they can be cited, and three estate files found this afternoon — Johan Casimir 1906 Smithfield, John Henry 1924 Ficksburg, George Augustus 1905 Bloemfontein. What was searched UNBLOCKED THE SAME EVENING: David signed in. The identical URL returned 3,499 again. Worth keeping rather than deleting, because the round trip is the controlled experiment — 3,499, then 0, then 3,499, one URL, one afternoon, one machine, and the only variable the session.
198. Done — finished and published this archive
Five of the Commandant's children are documented and deliberately not yet entered
Found 27 September in Dutch Reformed baptisms naming BOTH parents: GEORGE AUGUSTUS b. 30 Sep 1845, bapt. Colesberg 8 Mar 1846; PETRUS JACOBUS b. 4 May 1852; MARGARETHA ELIZABETH b. 16 Feb 1854; ANNA SUSANNA b. 15 Aug 1860, bapt. Bloemfontein; ANNA JACOBA b. 3 Nov 1864, bapt. Bloemfontein. Not entered, on purpose, for two reasons and the second is the real one. The arks were not captured before the session dropped, so they cannot be cited to the record. And four of the five collide with names already here — this archive holds five George Augustus Kolbes and two Petrus Jacobuses — so each needs a distinguishing byname decided on evidence rather than on convenience. This archive turned a widow into a daughter on 23 September by entering somebody quickly. What was searched 27 SEPTEMBER, LATER — DAVID SIGNED IN AND ALL FIVE ARE ENTERED. The identical URL returned 3,499 again, which is also the cleanest possible proof that the morning's 0 was the session and not the data. Arks captured and cited: George Augustus 1:1:6245-2S2X, Petrus Jacobus 1:1:6PYD-43CP, Margaretha Elizabeth 1:1:6P19-2Y5T, Anna Susanna 1:1:Q2HL-BMPQ, Anna Jacoba 1:1:Q2HL-Y1LG. Each is named `of 18xx` after the archive's own “Frederick Kolbe of 1816” convention, because four of the five collide with names already held. Nine of the twelve are now drawn in kin.js, eight of them on a register.
199. Next — ready to start this archive
A sixth child is probable and undocumented: Catharina Wilhelmina, who would be the eldest
The Free State registers have CATHARINA WILHELMINA KOLBE marrying JOHANNES PETRUS PIENAAR at Bloemfontein on 8 October 1860, born 1843 — the year the Commandant married, which would make her the eldest of the twelve. The record does not name her parents, so she is an inference from a year, a town and a surname. The Geni list's *Catharina Wilhelmina Magdalena Pienaar Anderson* is explained by it — Pienaar first, Anderson second — and a tree that explains an inference is not evidence for it. Her baptism, or her own death notice, would settle it.
200. Done — finished and published this archive
Row numbers in this file were integers except one, which was a string
Row 196's `n` was the STRING `"196"` where all 199 others were integers. It broke a `max()` here within minutes of being written, and it would have broken any sort, any lookup by row number and any gate comparing them. FIXED HERE RATHER THAN LEFT FOR ITS AUTHOR, and the reasoning is worth the line. The estate's habit — and this session's own rule — is not to touch another session's work, because commits in this tree have swept other people's changes in and out all month. **But the row is owned by *this archive*, it is `done` so nobody is still editing it, and the change is a type coercion that alters no meaning. Leaving a known landmine in shared data out of authorship etiquette is the worse error. A check refuses the next one**, so the fix is not a one-off.
201. Next — ready to start this archive
Ten Mountjoy estates at the Cape, and every one of them would name a father
Found 27 September in the new NAAIRS database, on the first sweep of the surname ever run. Seven Mountjoy daughters filed under married surnames — BRUMMER, DIESEL, DEMMER, BURGER, NEPGEN, FOURIE, KLOPPER — and three Mountjoy wives under Mountjoy. A death notice names parents, so each is a chance to attach a Mountjoy woman to a father, and the married-surname filing is the exact trap that hid the Commandant's widow. None is ordered and none should be ordered singly: they are almost all Cape Town Archives Repository, so they are one enquiry. References at /searched/.
202. Next — ready to start this archive
A Daschner child's estate this archive did not know about — TAB MHG 10287/69
DASCHNER, HEINRICH, with the archivist's note “CHILD.” This archive holds no Heinrich Daschner. Joseph Heinrich Daschner died in 1942 and his death notice is in custody here; a child's estate under 1969 is more likely a grandchild. A child's estate exists when a child owned something, usually an inheritance, so the file may name the parent it came from.
203. Blocked — waiting on a person or an institution National Archives of South Africa
The education-series sweep still cannot be run: old NAAIRS is 503
Row 1 wanted the Transvaal education series (source OD) swept for BARRY, MOUNTJOY, DASCHNER and KOLBE. The series sweep needs the old NAAIRS, which answers 503 Service Unavailable — twelve days after it answered «the connection to the Search Manager Server is inactive». The new database was swept instead and produced 58 Mountjoys and 12 Daschners, but NARSSA states in writing that it holds only part of 8.3 million migrating entries, so it cannot answer a question about a whole series. This row is about the old service coming back and nothing else. CORROBORATED INDEPENDENTLY AND THE ADVICE IS NOT TO RETRY. The Luwinski session probed the same endpoints and found the old Search Manager at 503 and `naairs.htm` at 404, and quotes NARSSA's page verbatim. Since the old database is down and the new one cannot clear a nil by definition, no instrument exists tonight that could lift a NAAIRS negative — so this is recorded once with dates rather than retried. A retry loop against an instrument that cannot answer is the most expensive way to learn nothing, and it manufactures fresh zeros.
204. Blocked — waiting on a person or an institution David
NARSSA's tariffs changed on 1 September and the reading room is closed
The Pretoria order is held on cost against figures that are now a month old. NARSSA posted new tariffs effective 1 September 2026 and asks researchers to read the schedule before submitting chargeable requests. Separately the reading room is closed — one notice for a temporary water-supply interruption, one headed *temporal closure of the reading room*. Nothing to do until David wants to revisit the cost, and worth knowing before anybody is asked to go in.
205. Next — ready to start this archive, with David
Sweep David's own mail — it is an unsearched repository and it has already produced a twelve-year tree
Found by David remembering, not by anything here. A 2008 letter carrying a 1,604-person descendant report of the English Barrys had been in the inbox for eighteen years and answered a lead this archive had marked *unknown and must not be assumed*. If one such letter is there, others may be. Worth searching for the surnames this archive works — Booyzen, Kolbe, Mountjoy, Daschner, Barry, Nepgen — and for correspondence from genealogists. A repository nobody has swept is exactly what /searched/ exists to record, and this one has never appeared on it.
206. Blocked — waiting on a person or an institution David
Ask Wendy for the IRISH Barry tree — she finished it, and did not send it
She sent one tree and had two. The 2008 attachment is *Descendants of Thomas Barry* of Stockton, Coventry — the English line, Hitchin to Swellendam. Ireland appears nowhere in it. Her covering note says she is “from the Irish mob who came out to South Africa in the early 1800” and had completed both trees. THE IRISH ONE IS THE ONE THIS FAMILY'S LORE POINTS AT — the Barry page quotes the family: *“As an Irishman, John Barry was reputed to be anti English”*. She also said she was about to start on “the other Barry family that came to South Africa in the late 1700 from London”, and eighteen years is long enough to have finished it. So there are two things to ask for and they are precise. THE 2008 ADDRESS IS ALMOST CERTAINLY DEAD — `telkomsa.net`, a service since wound down — so a bounce would say nothing about her; the guest farm's own website is the second route. Her contact details are deliberately not printed on this site: she is a living private person who wrote a personal letter, and they are held in the project's unpublished notes.
207. Next — ready to start this archive
The third Barry group — from London, late 1700s — is now a named direction
The first positive direction this archive has ever had for the Barry origin. Its wall is John Barry, died at de Bruins Poort 1853, with no birth, no arrival and no burial, and his son John Augustus Barry I, born 1819 “at sea”. If the family came from London in the late 1700s, the places to look are the London parish registers and the Cape arrival and burgher records of the 1780s–1800s — not the 1820 Settler papers, which is where the family story has always pointed and where the Kolbe page already says he was not. NOT ADOPTED: one researcher's sorting of a surname into three groups is a hypothesis.
208. Next — ready to start this archive
Order the Colesberg marriage register for 23 February 1846 — it would make the date documented
The date is adopted on arithmetic and a book, and it should rest on a page. Pama gives John Barry × Margaretha Wilhelmina Elizabeth Kolbe, Colesberg, 23 February 1846. Colesberg is a parish this archive already works — the Settler's death notice was signed there, his son George Augustus was baptised there in 1846, and Johan Casimir took membership and called banns there. The same register run, the same year. It would also give the marriage's witnesses, and this family's witnesses have twice turned out to be relatives.
209. Next — ready to start this archive
Test Pama's Burgersdorp-to-Dordrecht move against the baptism registers
Pama says the children were *born in Burgersdorp up to Nicolaas who, with his later siblings, was born in Dordrecht* — placing the household's move between 1855 and 1857. This archive already holds baptisms for most of these children and can test it without ordering anything. If it holds, it dates a migration from a book and confirms it from fonts; if it fails, it is one more reason to treat Pama as a lead list.
Nothing in this archive is in that state.
The other work lists
Each archive keeps its own, in the same shape. Nobody has a list of everybody — that is the point: the work sits with whoever can do it.
What does not go here
Findings do not. This archive already has five registers for those and they are better at it: what has been read holds every query ever run, negatives at the same length as finds; open questions holds what is not known, with the record set that would answer each; errands holds what can only be got by standing in a reading room; corrections holds every claim withdrawn and the cause; and what changed holds the diary.
None of them answered what is being done, and so for weeks that answer lived in a chat window — which is the one place that does not survive the session. That is the whole reason for this page, and it is the same failure this archive keeps finding in its sources: the material survives and the index to it does not.