Corrections

was That ARTURO LERENA was a doctor, published on his row and in a commit saying the Lerena Traibel were 'two professional men in one generation' - the occupation field read simply 'doctor', which in English means a physician.

now HE IS AN ABOGADO. His daughter's marriage act of 9 August 1909 gives him as 'de profesion ABOGADO'. In Uruguay DOCTOR IS THE STYLE OF A LAWYER as much as of a medico, and the 1880 marriage act's 'el Doctor Don Arturo Lerena' carries no more than the title. THE SECOND DOCTOR IS AFFECTED TOO: 'Dr Ambrosio LERENA' keeps the title and loses the implied profession, because nothing on the 1880 act says which he followed.

how BY READING THE NEXT ACT, WHICH IS THE ONLY WAY THIS KIND OF ERROR EVER SURFACES. Nothing was wrong with the transcription; the word DOCTOR was carried into English where it means something narrower. A TITLE IS NOT AN OCCUPATION, AND A TITLE TRANSLATED IS A CLAIM.

was That ALEJANDRO LERENA married a BARBARA LERENA, and that their son CARMELO ANSELMO LERENA was born and died at San Jose de Mayo in April 1846. Published that morning as one of four new Lerena-married-a-Lerena couples, on three person pages and in a commit.

now THERE WAS NO MARRIAGE. BARBARA WAS AN ENSLAVED WOMAN IN HIS HOUSEHOLD. The plate of the burial of 28 April 1846 reads 'di sepultura en el angelorio al cadaver de CARMELO ANSELMO, HIJO NATURAL DE BARBARA, ESCLAVA DE D.N ALEJANDRO LERENA, murio ayer de enfermedad, de edad de SIETE DIAS'. The indexer read 'esclava de D.n Alejandro Lerena' as a parent pair, made Alejandro the father, and GAVE BARBARA THE SURNAME OF THE MAN WHO HELD HER. The child is 'hijo natural' with no father named and no surname of his own. ALL THREE ROWS ARE RENAMED TO WHAT THE ACT SAYS and their addresses redirect: she is BARBARA, he is CARMELO ANSELMO, and Alejandro is a man of San Jose de Mayo in 1846 and nothing more. AND THE COUNT PUBLISHED THAT MORNING IS WRONG BY ONE: the query returned thirty-eight couples with a Lerena father and a Lerena mother, and at least one of them is not a couple at all.

how BY OPENING THE ACT, WHICH IS THE WHOLE OF THE METHOD AND WAS THE FIRST THING DONE WITH IT. The four couples were entered index-level that morning, every row saying INDEX-LEVEL, NO ACT OPENED, and this was the first one opened. THE LESSON IS NOT 'INDEXES ARE WRONG' - THIS ARCHIVE KNEW THAT. It is that an index which must output a father and a mother will MANUFACTURE them from whatever the act supplies, and that what it supplied here was a sentence about ownership. A PARENT-PAIR QUERY CANNOT DISTINGUISH A MARRIAGE FROM A SLAVE-HOLDING, and this archive ran a worldwide one the same morning.

was That VITALIANO LERENA married an ANDREA LERENA - the second of the four new Lerena-married-a-Lerena couples, published that morning with a note calling Trinidad 'the town this archive has worked hardest'.

now HER NAME IS ANDREA ETCHEBERSE. The marriage act of her daughter, Santisima Trinidad, 26 March 1906, reads 'hija legitima de Don VITALIANO LERENA y de Dona ANDREA ETCHEBERSE' - and the surname is BROKEN ACROSS THE PAGE OPENING, 'ANDREA' at the foot of one page and 'ETCHEBERSE.' at the head of the next. The index gave her the father's surname instead. THAT IS THE SECOND TIME IN ONE MORNING - BARBARA at San Jose de Mayo was the first. VITALIANO LERENA IS REAL AND STAYS, with a daughter born about 1874 and married at Trinidad in 1906, in none of this archive's sweeps. Three further corrections off the same plate: the groom is ETCHEBERSE and not Echeberse, he was THIRTY-THREE and not the index's three, and the priest STRUCK OUT 'LEGITIMO' AND WROTE 'NATURAL' over his parentage.

how BY OPENING THE SECOND OF FOUR ACTS, HAVING JUST BEEN BURNED BY THE FIRST. Both failures have one mechanism and it is now in data/false-zeros.tsv: an index with a MOTHER field will fill it from the nearest surname when the act does not fill it the way the schema expects. WHERE BOTH PARENTS SHARE A SURNAME, AN INDEX ENTRY IS MORE SUSPECT AND NOT LESS - that is the shape this failure produces.

was That the parent-pair query had produced FOUR URUGUAYAN LERENA-MARRIED-A-LERENA COUPLES this archive had never met - published on open question 1, in searched.json and in a commit, with fourteen people entered on the strength of it.

now ALL FOUR ARE NOW OPENED AND NOT ONE IS A DEMONSTRATED COUPLE OF TWO LERENAS. SAN JOSE DE MAYO, 1846: 'hijo natural de BARBARA, ESCLAVA DE D.n ALEJANDRO LERENA' - an enslaved woman given her holder's surname, and no marriage at all. TRINIDAD, 1906: 'hija legitima de Don VITALIANO LERENA y de Dona ANDREA ETCHEBERSE', the mother's surname broken across the page opening and replaced by the father's. MONTEVIDEO, 1865: 'hija de ANTONIO y de SABINA SERENA' - ONE surname between the two parents, which the index duplicated into two. CERRO LARGO/MONTEVIDEO, 1938: 'FLORENCIO SERENA, fallecido... y de TERESA SERENA, DE ESTADO VIUDA' - both named, but she is a widow and a widow is routinely written under her late husband's name. TWO DISPROVED OUTRIGHT, TWO UNPROVED.

how BY OPENING ALL FOUR ACTS, ONE AFTER ANOTHER, IN A SINGLE SITTING. Each row had said INDEX-LEVEL, NO ACT OPENED, which is why this cost a morning and not a reputation. THE SIX LERENA-MARRIED-A-LERENA COUPLES THIS ARCHIVE HELD BEFORE THE QUERY ARE UNAFFECTED - the two Trinidad couples, the 1911 dispensation, Alejandro, Eloy x Felisa and the Navarrete household all rest on ACTS, not on an index's parent fields. SO THE PATTERN STANDS AND THE QUERY THAT SEEMED TO MULTIPLY IT DOES NOT.

was That a SERENA sweep of Flores returned 290 RECORDS and that 190 of them remained unread - published in searched.json, in a worklist row and in a commit, an hour after the finding.

now THE FLORES SET IS ABOUT TWENTY-TWO RECORDS AND THEY WERE ALL READ. The URL parameter q.anyPlace RANKS RESULTS, IT DOES NOT FILTER THEM: the search returns every SERENA in Uruguay - an ordinary Italian surname there - sorted so the Flores ones come first. Page one carried 22 Flores records; pages two and three carried ZERO BETWEEN THEM. So '290' is the size of the Serena surname in Uruguay, not the size of the seam, and the 190 'unread' records are other families in other departments. THE FINDINGS THEMSELVES ARE UNAFFECTED - the children of Custodio and Ema, of Juan Bernardo and Eduviges, and the rest were all on page one and are all still there.

how BY PAGING TO THE END INSTEAD OF STOPPING AT A COUNT, WHICH IS THE SAME LESSON AS THE HERENCIA ONE AND WAS LEARNED THE SAME WEEK. On 11 October this archive wrote: 'a result count is not a finding... every hit has to be read before it is counted.' IT THEN REPORTED A COUNT AS A FINDING THE NEXT DAY. The difference is that a place parameter LOOKS like a filter, and the only way to know is to page to where the matches stop.

was That the gap in the dispensation sweep was a new finding - published yesterday on /method/ as 's-adjacency', in data/dispensations-compound-sweep.tsv and in a commit message, with the framing that the errand's premise was 'half wrong' and that this session had worked out which half.

now THE ARCHIVE HAD WRITTEN IT DOWN ON 14 SEPTEMBER, IN data/compound-surname-sweep.tsv, IN ALMOST THE SAME WORDS - 'its formula was LERENA FELIGRESA and the compound surname sits between the two words.' That file had already run the compound forms, already found the ninth dispensation (Jose Pedro Massera's petition of 19 June 1913 to marry Dona Ema Sixta LERENA JUANICO), and already drew the working rule from it. Nothing published yesterday about the mechanism was new, and the errand it was written against was STALE - the work had been done and the errand never closed.

how By doing the work properly today and walking into the archive's own record of it. Opening the 1913 dispensation plate to check whether it was new led to a grep for Massera, which returned the September file. The published claim was caught by the only thing that catches this class of error - finishing the job rather than describing it.

was That LEANDRO LORENA and his wife LLANOS are a true null - 'the army card is their only record in the index, in any country.' Published 10 October on open question 1, in data/cajon-985-last-two.tsv and data/cajon-985-origins.tsv, and offered as this archive's model of what a properly controlled null looks like.

now THEY ARE ON THE 1895 BUENOS AIRES CENSUS, lines 7 to 11 of Seccion 18 Subdivision 34, read at the plate on 12 October: LERENA Leandro, 38, espanol, born SEGOVIA, literate, F.te de Aceite; LLANOS Lucia, 32; Emilia 9 (Spain); Francisco 5 (La Plata); Paulina 8 meses (portena). The search itself was sound - signed in, constraint binding, eleven Llanos rows returned. A SPOUSE-PAIR QUERY IS STRUCTURALLY BLIND TO A CENSUS, which records no marriage relationship to constrain on; and the card spells him LORENA where the census spells him LERENA.

how By opening the census sheet, which FOUR of the five people were already in this register from. The register called them 'a separate, Spanish-born family' while open question 1 called the same man a true null, and neither row looked at the other. The missing link was the wife: LUCIA LLANOS had never been entered, and she is the only name the two records share.

was That the department of ARTIGAS holds 208 LERENA records, against 699 for Flores - published this afternoon in data/artigas-opened.tsv, on /uruguay/#s-artigas, in searched.json and in a commit, with the framing that an unworked department held a third as many records of the name as the heartland.

now THE PLACE FILTER FALLS BACK TO THE COUNTRY, so those counts are not departmental. q.anyPlace='Nonesuch, Uruguay' - a department that does not exist - returns 1,787, which is exactly the whole-country figure. Real department names do bind (Artigas 208, Flores 699, Montevideo 891), but a record whose indexed place is only 'Uruguay', or which carries no place at all, matches every departmental query. Of the first hundred of the 208: 15 actually name Artigas, 28 name somewhere else entirely - Montevideo, Buenos Aires, the United States - and 57 carry no place in the index at all.

how By opening a record the Artigas query returned and finding it was a MONTEVIDEO death. The 1896 death of 'Pedro a Lerena' came back under Artigas; its volume is 'Montevideo. Death Records 1896'. The control that was run - Osaka, Japan returning 0 - proved the filter binds on COUNTRY. It never tested whether it binds on DEPARTMENT, which is the thing the claim rested on. A control licenses the shape of a query, not the value it returns.

was That the Cordon parish holds 470 documents of this surname and is 'the largest unworked seam in the archive'. Published this morning in data/cordon-sized.tsv, on /uruguay/#s-cordon, in searched.json, in an errand and on the work list - with the warning, in the same file, that an errand understating its own size is how a seam stays unopened for a month.

now THE PARISH'S OWN SHARE IS ABOUT 95. q.anyPlace=Cordon scopes the NEIGHBOURHOOD: 95 of the first 100 hits are Nuestra Senora del Carmen La Mayor volumes and pages 2 and 3 contain not one - the rest are La Union, Pocitos, San Francisco de Asis, the Catedral and the Montevideo civil marriage registers, whose people merely lived in the Cordon. So the file overstated its own size by five times, in the same breath as warning against understating it.

how By paging the 470 to the end instead of trusting page one - the same check that was skipped at Artigas this afternoon, and skipped here this morning. Controlled properly this time: a fake sub-place, 'Nonesuch, Montevideo', returns 26 pages (the whole department) while Cordon returns 5, so the filter does resolve Cordon. THIS IS THE SECOND PLACE-FILTER ERROR OF THE DAY AND THEY HAVE DIFFERENT CAUSES: at Artigas the filter fell back to the country; here it resolved correctly and the ARCHIVE read a barrio as a parish.

was That Montevideo household 229 is a census of about 1833, and that its head is PEDRO ANTONIO LERENA, SON OF SANCHO, born 31 July 1793. The c.1833 has been carried in data/lerena-register.tsv for weeks and three birth years are derived from it - Emilia LERENA Vega c.1815, Pedro LERENA Vega c.1817, Vicenta LERENA Vega c.1821.

now THE ENUMERATION IS THE PADRON OF 1850, CONFIRMED AT CATALOGUE LEVEL. FamilySearch catalogue koha:1408224, 'Uruguay, Montevideo, censos, 1726-1860', lists film 7499042 item by item: item 1 = Padrones de habitantes, LIBRO 263, 1843; item 2 = LIBRO 264, 1850; item 3 = libro 257, 1852-1853. The household sits in item 2, as the viewer's own breadcrumb says. So the three birth years are each seventeen years too early: Emilia is about 1832, Pedro about 1834, Vicenta about 1838. And the head, entered at forty, was born about 1810 - which is not a man born in 1793.

how By opening the plate to settle a conflict between the 1896 death act and the census age, and then going to the catalogue rather than stopping at the volume label. The c.1833 was itself an inference from the head's age matching Sancho's son, so the error was circular: the date was chosen to fit the identification, and the identification then rested on the date. Under EITHER candidate book - 1843 or 1850 - the head is too young to be the man born in 1793, so the identification does not survive the correction whichever way the volume falls.

was That CARMELO ANSELMO was born on 20 April 1846. Entered from the burial act, which gives only an age at death.

now The baptism act of 25 April 1846 states it: 'que nacio el VEINTIUNO del mismo' - 21 APRIL. It also squares with the burial's 'de edad de siete dias' on the 28th.

how By finding the baptism three days earlier in the same parish's books, through a phrase search that had failed an hour before because of an abbreviation.

was That the surname of this household was ZALSAMENDI - carried since 21 September 2026 on five register rows (Eduardo ZALSAMENDI, Juan ZALSAMENDI, Dolores POU de ZALSAMENDI, Maria Margarita ZALSAMENDI Lerena, and Amelia LERENA 'm. Eduardo ZALSAMENDI'), and RE-ASSERTED ON 11 OCTOBER 2026 IN A COMMIT THAT ARGUED FOR IT: when the civil index spelled the man ZALDUANDO, this session reasoned that a PLATE outranks an INDEX and published ZALSAMENDI as the better reading.

now THE SURNAME IS ZALDUONDO, AND BOTH PLATES SAY SO. Read on 11 October 2026 at full resolution: the 1903 Pocitos act heads the entry 'Maria Zalduondo', gives the father as 'Eduardo Zalduondo' and the paternal grandmother as 'Dolores Pou de Zalduondo'; the 1913 Montevideo death act, certificate 553, has 'ha fallecido Eduardo Zalduando' and 'era hijo de Juan Zalduondo y de Dolores Pou, fallecidos'. ZALSAMENDI APPEARS ON NEITHER PLATE. It was a misreading of the 1903 hand, and it stood for three weeks.

how BY READING THE PLATE INSTEAD OF REASONING ABOUT IT. The argument that convicted the index was itself the error: 'a plate outranks an index' is true, but THIS SESSION WAS NOT COMPARING A PLATE TO AN INDEX - IT WAS COMPARING ANOTHER SESSION'S READING OF A PLATE TO AN INDEX, and treating the two as the same thing. The index had been right all along, and a further tell was ignored: the same index gave the child a fourth name, LUISA, which the 1903 reading had dropped and which the plate does carry. THIS IS THE LENDIN LESSON AGAIN - a reading is only as good as the smallest thing it rests on - and the archive had even written the warning down, noting that a MACHINE transcript of this same act read 'DEL SUANDO'. Two readers had already failed on this hand; that was evidence the hand was hard, and it was used instead as evidence the index was careless.

was That the open question was whether to believe a PLATE or an INDEX about the wife's forename - published on 11 October 2026 as open question 10, 'Amelia or Angelica?', which framed the 1903 reading as plate-authority against index-error and said the test was to open the 1913 act.

now THE TEST WAS RUN AND THE FRAMING WAS WRONG: IT IS PLATE AGAINST PLATE. The 1913 death act is explicit - 'era casado con ANGELICA LERENA, de cuya union quedan tres hijos llamados: Maria Elena, Eduardo Avelino y Maria Margarita' - and the 1903 baptism is equally explicit that Maria Margarita Luisa is the daughter of Eduardo Zalduondo and AMELIA LERENA. TWO IMAGES, TWO FORENAMES, ONE CHILD ATTRIBUTED TO BOTH - which is what forces them to be one woman rather than two sisters, because the 1913 act places Maria Margarita inside the union with Angelica and the 1903 act places her inside the union with Amelia.

how BY THE USER SIGNING THE SESSION IN, AND BY READING BOTH PLATES RATHER THAN ONE. The question had been filed as blocked the same morning. THE CORRECTION IS NOT THAT THE QUESTION WAS ASKED, BUT THAT IT WAS ASKED WITH A THUMB ON THE SCALE - it assumed the unread document would vindicate the read one.

was That DOLORES POU was 'Catalan or Balearic', and that her household was 'Spanish on both sides a generation back, as the Lerena were'. Written on her row and on her husband's on 21 September 2026, reasoned from the surname alone.

now SHE IS BRAZILIAN. Her son's marriage act of 20 November 1897, read from the plate, gives 'Dona DOLORES POU, de 58 anos ... de nacionalidad BRASILENA'. And the same act gives her husband JUAN ZALDUONDO as 'de nacionalidad ORIENTAL' - Uruguayan-born, aged eighty, not a recent Spanish arrival either. THE SURNAMES MAY WELL BE CATALAN AND BASQUE; THE PEOPLE WERE NOT.

how BY READING AN ACT THAT ASKED THE QUESTION DIRECTLY. A Uruguayan civil marriage form has a 'nacionalidad' blank for each parent, so it answers in one word what a surname can only suggest. THIS IS THE ARCHIVE'S OWN RULE - GO WHERE THE FORM ASKS THE QUESTION - AND IT WAS NOT APPLIED FOR THREE WEEKS because nobody had looked for the marriage. The error was not the guess, which was labelled a guess; it was letting a labelled guess stand while the document that settles it went unsought.

was That the MORATORIO surname recurring in this register was 'a coincidence of surname in one city and it is not evidence of anything' - written on Eduardo Zalduondo's row earlier the same day, after noticing that the Juez de Paz at his 1913 death was an Eduardo J. Moratorio.

now THE CAUTION WAS RIGHT ABOUT THE JUEZ DE PAZ AND WRONG ABOUT THE FAMILY. There IS a Moratorio connection and it is close: FERNANDO MORATORIO, fifty, of commerce, WITNESSED ANGELICA'S MARRIAGE on 20 November 1897, and this register already held a Fernando Moratorio as the husband of JOSEFA LERENA TRAIBEL, the bride's own aunt. NOTHING IS CLAIMED ABOUT THE JUEZ DE PAZ, who remains unconnected.

how BY FINDING THE MARRIAGE ACT FOUR HOURS LATER. The caution was written to stop a future reader inventing a link from a shared surname; it had the side effect of asserting there was none. 'NO CONNECTION IS CLAIMED' AND 'THERE IS NO CONNECTION' ARE DIFFERENT SENTENCES and the row used the first to mean the second.

was That SOCORRO LERENA's marriage is dated 15 JULY 1885. Written on her register row earlier the same day from a parish index row reading 'Socorro Lerena, Bride, Marriage 15 July 1885, Spouses Serapio Fernandez'.

now SHE IS THE GROOM'S MOTHER, NOT HIS BRIDE. This register already had her born about 1832, a widow and a property owner in 1897, and the mother of the Serapio Fernandez who married a Lerena in 1885. The index record gives its bride as thirty, soltera, born 1855. A woman cannot be both. THE MARRIAGE IS ALMOST CERTAINLY ANTONINA LERENA'S, dispensed on 26 June 1885 for a second grade of consanguinity, three weeks before.

how BY OPENING THE INDEX RECORD INSTEAD OF READING THE SEARCH ROW. The search row gave a name, a date and a spouse and looked complete. The record behind it shows the bride carrying THE GROOM'S OWN ATTRIBUTES, FIELD FOR FIELD - same age, same birth year, same birthplace, same residence, same marital status. THAT IS A PARSER DUPLICATING ONE PARTY ONTO A NAME IT FOUND ELSEWHERE ON THE PAGE, and the name it found was the mother's. THE RULE THIS YIELDS: WHEN TWO PARTIES ON AN INDEXED MARRIAGE SHARE EVERY ATTRIBUTE, ONE OF THEM IS NOT A PARTY.

was That this archive was making every person it found into a person. Not written anywhere as a claim - which is the point. It was the unstated assumption behind a day's work, and behind a gate, tools/check-people-named.py, that reported 'every person named on a record is in the register' while the day's findings went into prose.

now 39 OF 45 PEOPLE NAMED IN ONE SESSION'S WORK HAD NO ROW. Measured on 11 October 2026 when David asked the question for the THIRD time. Among them: VIOLETA BLANCA LERENA, born 29 October 1915, whose existence is this archive's published argument that the 1911 dispensation was granted - a keystone of a page, and unclickable. AND ELEVEN OTHER TRINIDAD CHILDREN with birth dates and named parents, published since 27 September. AND EDUARDO ZALDUONDO'S OWN BROTHER, with a death date, a birth year, both parents and a wife - left out on a day when this session created a row for a godmother about whom nothing whatever is known.

how BY BEING ASKED, AND THEN BY MEASURING INSTEAD OF ANSWERING. The gate said green. The gate watched FIVE files and every one of them was TAYLOR, BRAYLEY or SOUTH AFRICAN - IT WAS BLIND TO THE LARGER HALF OF THE REGISTER, which is Uruguayan. Its own docstring says 'the gate grows by being made to', and it had been grown once before, on 8 October, when David asked a version of this same question. THE FIX IS NOT THE TWENTY-ONE PEOPLE ADDED; IT IS THAT THE GATE NOW WATCHES THE URUGUAYAN FILE TOO, so the next session cannot repeat this quietly. THE REAL FAULT WAS INCONSISTENCY, NOT OMISSION: a Juez de Paz and a witness were admitted the same afternoon a documented brother was not.

was That Don Clemente GARCIA died 'by 1800' - the name of his row in this register, and the date this archive has carried since it first read the probate.

now HE WAS ALREADY DEAD IN 1794. The earlier file on the same estate opens 'caminamos a la ESTANCIA DEL FINADO Don Clemente Garcia' and calls Celestina Gonzalez 'LA VIUDA' - six years before the case this archive had read. So the 1800 court order is not the settling of a fresh death but a SECOND reckoning of an estate already inventoried once, reopened because the widow had remarried and there was a minor child. THE ROW IS RENAMED TO 'd. by 1794' AND HIS ADDRESS DOES NOT MOVE. I nearly published the opposite - that the date was stuck in his URL and the row had to carry a wrong title forever - and then read slugify(), which strips parentheses. THE DISAMBIGUATOR IN A ROW TITLE IS NOT PART OF THE ADDRESS, so dates in brackets can be corrected freely; only the frozen ledger key has to be moved by hand, and it was.

how By paging further into the same full-text result set instead of stopping at the first hit. The 1794 file sat 20 results below the 1800 one and no search would have surfaced it as new - it is the same case name, the same court, the same man. THE LESSON IS THE CHEAPEST ONE IN THIS WHOLE ARCHIVE: when a result set has 700 entries and you have read 20, the next 20 are not a long shot. THE SECOND HALF OF IT WAS CAUGHT BY NOT TRUSTING MY OWN REASON FOR GIVING UP. Writing 'the row cannot be renamed' took ten seconds; checking whether that was true took one grep of tools/build-people.py and showed it was not.

was That the Navarrete household was the proof that a Lerena could marry a Lerena - published on /method/, /uruguay/, /argentina/, /name/ and open question 1, which is the page this whole archive is organised around.

now IT IS THE SIXTH SUCH COUPLE, NOT THE FIRST. The register already held two Trinidad couples (1894, 1913), the double-first-cousin dispensation of 1911, Alejandro, and - found the SAME MORNING, written into the SAME TABLE a few hours earlier - ELOY LERENA x FELISA LERENA, baptising their daughter Juana at San Jose de Flores on 10 November 1912, both 'natural de Espana'. His own row says in capitals that he 'MAKES A FIFTH LERENA-MARRIED-A-LERENA COUPLE'. SO THE OBJECTION I CLAIMED TO HAVE WEAKENED HAD ALREADY BEEN ANSWERED, BY ME, BEFORE LUNCH. What Navarrete really adds survives the correction and is worth having: the first such household outside South America, the first where the children carry the name doubled as a surname of record, and the first place on a map where the practice can be searched.

how BY A TOOL AUDIT THAT HAD NOTHING TO DO WITH IT. Rewriting tools/last-alive.py threw up 'Eloy LERENA (natural de Espana, m. Felisa LERENA)' as a row my own audit regex mis-flagged, and reading it to clear the flag showed the phrase 'HIS WIFE IS ALSO A LERENA, which this register has now met five times'. NOTHING IN THE GATES COULD HAVE CAUGHT THIS. A claim of novelty is not a claim about a person, a date or a link, so no consistency check touches it - the only defence is to search this register for the thing you are about to call new, and I did not.

was That Avelino Lerena Fernandez was the GREAT-GRANDFATHER of Angel Miguel Avelino Lerena Larriera, published on his person page since 21 September.

now HE IS THE GRANDFATHER. Angel's father is Pedro Alberto Ysabelino Lerena Traibel; Avelino is that man's father. One step, not two. THE EDGES IN data/relations.tsv WERE RIGHT THE WHOLE TIME - father and grandfather are both there and both correct - so the chart on the page disagreed with the prose beneath it, and nothing checked that they agree.

how BY LEARNING THE OTHER GRANDFATHER'S NAME. The 1879 christening of his sister names the maternal grandfather MIGUEL LARRIERA, which makes ANGEL MIGUEL AVELINO a child named for BOTH grandfathers - and reading the name that way made the generation count obvious. A NAMING PATTERN IS EVIDENCE ONLY WHEN YOU KNOW BOTH SIDES OF IT; with one side it is a story.

was That DON DOMINGO RODRIGUEZ (the elder, of la Colonia) was a person - published at /people/don-domingo-rodriguez since 21 September, drawn on his grandson's family chart, and given a wife.

now HE NEVER EXISTED. The 1840 Catedral act was quoted on this archive's own page as 'Nieto pat[erno de D. DOMINGO] RODRIG.z' - and the forename inside those square brackets was supplied by the reader, almost certainly because the CHILD being baptised is called Domingo. THE PLATE READS JOAQUIN: 'Nieto Pat.o de D. JOAQUIN RODRIG.z y D.a Juana del Espiritu S.to: ambos nat.s de la Colonia del Sacram.to', and the act's own names panel independently gives 'Pats. Joaquin Rodriguez'. The paternal grandfather is MAYOR DON JOAQUIN RUIZ DE CARVALLO, already in this register, named on the other grandchild's act of 1827 and under this register's known second spelling of the surname. AND THE SAME BRACKET SPLIT ONE WOMAN INTO TWO: 'Dona Juana del ESPIRITU SANTO' was entered as the phantom's wife, while 'Dona Juana SUAREZ' sat in the register as Joaquin's. Both are the mother of Capitan Antonio; both rows were the same woman, and they are merged. TWO PUBLISHED ADDRESSES NOW REDIRECT.

how BY A SITE-WIDE AUDIT, AND ONLY BECAUSE A SOFT WARNING WAS READ INSTEAD OF SKIMMED. tools/consistency.py had been listing 'Dona Juana del ESPIRITU SANTO and Dona Juana SUAREZ may be one person entered twice' among eight things to look at. Three of those eight were a bug in the gate itself - it compared a person with herself - and noise is what let the real one sit. THE LESSON IS ABOUT SQUARE BRACKETS: this archive uses them for honest editorial reconstruction of damaged text, and that is right, but a reconstruction inside a quotation reads exactly like a transcription once it is three weeks old. A NAME SUPPLIED BY A READER MUST NEVER BECOME A PERSON ROW.

was That the wife of Eduardo Lerena was MARIA MERCEDES ROZZETTI, published at /people/maria-mercedes-rozzetti/ with her daughter at /people/ana-maria-lerena-rozzetti/.

now THE PLATE SAYS RAZZETTI. Both rows were built from the FamilySearch index of a 1921 baptism. The act itself, at Nuestra Senora de los Desamparados in the partido de Adolfo Alsina, writes the surname twice on one line in a clear hand - 'de D.a MARIA MERCEDES RAZZETTI' and, four words later, 'Padrinos: D. DOMINGO RAZZETTI' - so the reading is confirmed by the same scribe in the same sentence. THE INDEX ALSO PUTS THE EVENT IN THE WRONG PLACE, normalising it to 'Buenos Aires City' while its own original-place field says ADOLFO ALSINA and the act says 'domiciliados en ESTE PARTIDO' - a rural district of the province four hundred kilometres away. Both addresses redirect.

how BY CHASING A DUPLICATE, NOT A SPELLING. This register holds an EDUARDO LERENA dispensed at Montevideo in December 1919 to marry a MARIA MERCEDES RAZZELLO, and the two rows had never been compared. Opening the daughter's plate to test that pair produced the correction instead - and made the pair stronger, because RAZZELLO and RAZZETTI are one hand read twice while ROZZETTI was never on the page at all.

was That this register held two men called Eduardo Lerena - one married at Montevideo in 1919, one a father at Buenos Aires in 1921 - kept apart that morning with the note that 'no act yet read says this man came from Las Piedras'.

now THEY ARE ONE MAN, and the act that proves it was found the same evening. The Canelones civil marriage of 11 January 1920 at Santa Lucia gives the bride's parents as DOMINGO RAZZETTI and JOSEFINA BERTOGLIO - and those two stood godparents to his daughter Ana Maria at Adolfo Alsina, Buenos Aires, on 19 January 1921. THE SAME FOUR PEOPLE IN TWO COUNTRIES THIRTEEN MONTHS APART. The rows are merged and /people/eduardo-lerena-2 redirects. THE SEPARATION WAS RIGHT WHEN IT WAS MADE, and is recorded here because the archive should be able to see how long one of its NOT MERGED notes survived: about four hours.

how BY FOLLOWING THE PAIR'S OWN 'WHAT WOULD SETTLE IT'. That line said: the Montevideo marriage act behind the 1919 dispensation, which will name his parents. It does - ARTURO LERENA and EMA REGULES - and it also gave him a birth date, 22 April 1886 at Montevideo. A PAIR ENTRY THAT SAYS WHAT WOULD SETTLE IT IS A RESEARCH INSTRUCTION, AND THIS ONE WAS FOLLOWED THE SAME DAY IT WAS WRITTEN.

was Three conclusions published on 10 October from a FamilySearch session that had silently SIGNED ITSELF OUT. (1) that LEANDRO LORENA and LLANOS are absent from the index, recorded as a CONTROLLED NULL; (2) that CONCEPCION LERENA is absent; and (3) a method finding that a parent-pair query and a spouse-pair query reach two different index shapes, built on a clean 0 Results for Carlos Lerena and Adela Carniglia. All three went into data/false-zeros.tsv, data/cajon-985-origins.tsv, /open-questions/ and a commit.

now THE SESSION WAS SIGNED OUT AND THE PAGE SAID SO IN ONE LINE AT THE TOP: 'Create a free account or sign in to see all available results.' A SIGNED-OUT SEARCH RETURNS A RESTRICTED, FUZZY TIER - twenty-one rows of United States census for every query, whatever the constraint. THE PROOF IS THE CONTROL ITSELF: the Carniglia spouse-pair query, which returned three true hits earlier the same day and from which two acts were read off the plate, now returns twenty-one rows and NOT ONE CARRYING THE NAME. SO THE NULLS ARE WITHDRAWN AND THE INDEX-SHAPE FINDING IS WITHDRAWN - the 0 it rested on has a simpler explanation.

how BY GOING BACK FOR ARTURO'S ENTRE RIOS MARRIAGE AND FINDING THE SAME URL RETURNING NOISE THAT HAD RETURNED TRUE HITS AN HOUR BEFORE. THIS ARCHIVE ALREADY HELD THIS TRAP, AS DISGUISE SEVENTEEN: 'THE SESSION THAT LOGGED ITSELF OUT AND DID NOT SAY SO'. THE CONTROL WAS RUN BEFORE THE NULL AND NOT AFTER IT, which is the whole lesson: on a service that can change state mid-session, a control is only evidence if it is re-run AFTER the null it is meant to protect. WHAT SURVIVES: everything read OFF A PLATE - Carlos Lerena natural de Montevideo, Arturo Lerena Garcia, the Valparaiso act - because those are transcriptions of images, not search results.

was That of the Lerena fathers in army drawer Cajon 985 whose origin can be traced, 'EVERY ONE is Uruguayan or married in Uruguay'. Published in data/cajon-985-origins.tsv since 13 September, strengthened on 9 October when CARLOS LERENA was added as a Montevidean, and leaned on in /open-questions/ question 1 as a BASE RATE - that an Argentine Lerena of this generation is a Uruguayan family's son far more often than not, and that Juan Carlos Lerena, absent from Rosario, fits that shape.

now ARTURO LERENA IS SPANISH AND HIS WIFE IS PERUVIAN. The marriage of his son Antonio at Crespo, Entre Rios, on 24 October 1924 - a printed form with a nationality blank for each parent - reads 'hijo de ARTURO LERENA, de nacionalidad ESPANOLA, de profesion PERIODISTA, y de AMALIA DIAZ, de nacionalidad PERUANA'. FIVE FATHERS ARE NOW TRACEABLE AND THE GENERALISATION FAILS ON THE FIFTH. What survives is the weaker statement: none of the five is Argentine. AND THE INFERENCE ABOUT JUAN CARLOS MUST WEAKEN TO MATCH - a drawer containing a Spaniard and a Peruvian is a drawer of immigrants, not a drawer of Uruguayans.

how BY WORKING THE ONE FATHER THE PATTERN HAD NOT YET BEEN TESTED ON. The claim was made on three men, restated on four, and broke on the fifth. THAT IS WHAT A BASE RATE BUILT ON A HANDFUL DOES, and this archive wrote it as a pattern rather than as a count. THE HONEST FORM WAS ALWAYS AVAILABLE: 'three of three', not 'every one'.

was That ENRIQUE CARLOS, son of CONCEPCION LERENA, was born at MENDOZA in 1899. Carried in data/cajon-985-origins.tsv since 13 September from the FamilySearch index, which gives his birthplace as 'Capital, Mendoza, Argentina'.

now THE CARD SAYS 'CAP. FEDERAL' - BUENOS AIRES. Read off the plate on 10 October 2026: 'LERENA ENRIQUE CARLOS... nacido en Cap. Federal'. THE INDEXER EXPANDED 'CAPITAL' INTO THE WRONG CAPITAL, and this archive copied it. TWO OTHER THINGS ON THE SAME CARD WERE ALSO INDEXED WRONG: the surname is LERENA, in his own signature, where the index says SERENA; and the index gives him a father, 'Spouses Carlos', where THE CARD'S FATHER FIELD IS STRUCK THROUGH WITH A PEN, the indexer having taken CARLOS from the son's own forename.

how BY OPENING THE CARD INSTEAD OF TRUSTING THE INDEX ROW THAT DESCRIBED IT. Three errors in one index entry, and this archive had copied one of them into its own file for four weeks. THE CARD IS A FINGERPRINT FICHA IN A DIFFERENT DRAWER FROM CAJON 985, which is why it had never been opened: it was reached through the person, not through the drawer.

was That ANTONINA LERENA the elder was born about 1830. Carried in the register since 9 October, inferred from her daughter Maria Pelaez being thirty at her marriage in 1880 - 'A DAUGHTER OF THIRTY IN 1880 PUTS HER OWN BIRTH NEAR 1830'. And that she was a natural de SAN JOSE, from the 1858 expediente.

now SHE WAS NINETY-TWO WHEN SHE DIED ON 22 JULY 1911, SO BORN ABOUT 1819 - eleven years earlier than the inference. And her death act says NACIDA EN MONTEVIDEO, not San Jose. A STATED AGE BEATS AN INFERRED ONE, and the Montevideo birth fits her father's own household: three of Ramon Lerena's children were christened at Montevideo between 1812 and 1821. THE SAN JOSE READING IS RECORDED AS A CONFLICT RATHER THAN DELETED, because the 1858 expediente plainly says it.

how BY A COMPOUND-SURNAME SEARCH AIMED AT SOMETHING ELSE. Work-list row 87 asked for 'Lerena Pelaez' to be crossed with the DISPENSATION series. It found no dispensations and four death and confirmation acts instead. THE INFERENCE WAS SOUND ARITHMETIC ON THE ONLY NUMBER AVAILABLE, AND IT WAS STILL ELEVEN YEARS OUT - which is what an inferred birth year is worth beside a stated one.

was That this register held TWO men - 'Alejandro LERENA (m. Elisa VILLARDEMOROS)' and 'Alejandro M. LERENA (with Julia BASSANI)'. They were proved the same man on 27 September 2026 and left as two rows for twelve days afterwards.

now ONE MAN, AND THE MERGE IS DONE. He is now Alejandro M. LERENA (m.1 Elisa VILLADEMOROS; m.2 Julia BASSANI), carrying both marriages, both families and both sets of evidence. WHAT PROVED IT, IN SEPTEMBER: the Cordon act of 26 December 1877, film 007713388 image 02455, names 'D. ALEJANDRO M. LERENA y ... D.a ELISA [VILLADEMOROS] (hoy finada)' as the parents of Jacinta Mercedes Gregoria and gives his own parents as 'D. AVELINO y D.a AMBROSIA TRAIBEL'. The 1886 act calls him viudo and Julia Bassani soltera, which is the second marriage. NINE RELATIONSHIP EDGES, a slug, the pair file and the three-Alejandros table all moved with him.

how IT WAS NOT CAUGHT - IT WAS OWED, AND THE ARCHIVE SAID SO IN WRITING. data/_person-pairs.txt carried the line: 'They are kept as two rows ONLY because a merge touches slugs, last-attested-alive.tsv, relations.tsv and the CHECKED_PAIRS in tools/consistency.py, and a half-done merge is worse than an honest pair. THIS LINE IS A DEBT, NOT A JUDGEMENT.' A debt recorded in those terms is the easiest thing in this archive to leave forever, and the hardest to defend leaving.

was That Alejandro M. Lerena's parents were 'STILL NOT NAMED BY ANY RECORD'. Carried in data/the-three-alejandros.tsv, under the heading of the question that table exists to answer, with a careful explanation of a blank line on an 1888 printed form.

now THEY WERE NAMED ON 27 SEPTEMBER 2026 - AVELINO LERENA and AMBROSIA TRAIBEL - on the Cordon act of 26 December 1877, read off film 007713388 image 02455. The cell went on saying 'STILL NOT NAMED BY ANY RECORD' for twelve days after the answer was found and written into data/_person-pairs.txt. The 1888 blank-form reasoning is kept below it, because it was true when written and explains why the question was hard.

how BY GOING BACK TO THE MERGE AND READING THE TABLE THAT DESCRIBES IT. This is the same failure as the Taylor strikes of 8 October: a finding is written where it is MADE and not carried back to every page that states the question. check-superseded.py would have caught it if the claim had been on its list; it was not, and a gate only knows what it is told.

was That FamilySearch's full-text corpus does not reach South Africa - 'a full-text search with a South African place filter errors outright and no South African collection surfaces, so a Cape question gains nothing here'. Published yesterday, in data/fulltext-route-2026-10-09.tsv and on /searched/.

now SOUTH AFRICA IS IN IT. The URL parameter errors; the in-page PLACE FACET works, and the surname returns SEVEN HITS across South Africa Legal 1559-2017, Orange Free State Legal, SOUTH AFRICA PROPERTIES 1800-1997 and Transvaal Religious. Two are this family and new: ROQUE LUIS ARMANDO LERENA on a Union voters' roll 1925-1931, polling district 770, and NUNO FERNANDO LERENA in a land record, identified by his exact birth date 6.10.1924. WHAT SURVIVES OF THE ORIGINAL CLAIM: there is no Cape civil registration in the corpus, so row 106 still gains nothing - but for a different reason than the one given.

how BY WORKING THE ROUTE INSTEAD OF TRUSTING YESTERDAY'S NOTE ABOUT IT. The error is a familiar shape in this archive and now has a fourth instance: A BROKEN PARAMETER READ AS AN ABSENT RECORD. data/false-zeros.tsv already holds three - FamilySearch's place filter returning 0 where the control proved thousands, its date filter zeroing a 413-result query, and findmypast's URL parameters being ignored entirely. The rule that would have caught it is the one this archive already wrote: run the control before believing the null. A search tool that ERRORS is not a corpus that is EMPTY.

was That the Montevideo MARRIAGE DISPENSATION RECORDS are worth opening because they name a groom's PARENTS. Written into work-list row 194 yesterday, quoting this archive's own canelones-films.tsv: 'the pre-marriage investigation is worth MORE than the act - it routinely gives a Spanish groom's PARENTS.'

now TRUE OF THE EXPEDIENTE, NOT OF THIS BOOK. The Arquidiocesis de Montevideo Marriage Dispensation Records are the Vicar General's SUMMARY LEDGER - one sentence per couple, recording whom he dispensed and from what. NOT ONE OF THE NINE LERENA ENTRIES READ TODAY NAMES A PARENT. The informaciones matrimoniales are a different series, and two record classes were conflated because both are called dispensations in English. AND THE LEDGER IS STILL WORTH MORE THAN EXPECTED, FOR A DIFFERENT REASON: a consanguinity dispensation states the DEGREE of kinship between the couple, which the Church computed before it would dispense - second degree equal lateral is first cousins, third is second cousins, and 'with attingency to the second' is two blood paths at once. That constrains a pedigree where no parent is written down.

how BY OPENING NINE OF THEM INSTEAD OF ONE. The first entry read, on 8 October, was a plain banns dispensation and said nothing either way; the expectation survived because it was never tested against a second.

was That ZELMIRA CARLOTA LERENA LARRIERA married Adolfo Cordero at Montevideo on 28 FEBRUARY 1901. Carried in her register row since 14 September, from an index.

now THE ACT SAYS 28 FEBRUARY 1904. Same day, same month, three years later. THE BOOK DECIDES IT: the register is La Union, Montevideo, Religious Marriage Records 1902-1934, so a 1901 marriage cannot be in it, and the entry immediately above this one is dated May 1902. AND IT MAKES SENSE OF THE DISPENSATION: the Vicar General lifted their consanguinity impediment on 1 February 1901 - they were cleared to marry three years before they did. A SECOND DISCREPANCY IS RECORDED AND NOT RESOLVED: the act makes her twenty in February 1904, which means born about 1883-84, where this register has her born 4 November 1880. Both stand.

how BY FULL-TEXT SEARCH ON HER FORENAMES, after the dispensation ledger turned out to name no parents. The ledger entry found yesterday gave the impediment and the date it was lifted; the act gives four parents, two ages, two nationalities, a profession and an address, and confirms the ledger's wording of the impediment word for word.

was That JUAN BERNARDO LERENA PELAEZ and the JUAN LERENA of the 1894 dispensation were DIFFERENT PEOPLE. Declared in data/_person-pairs.txt earlier the same day: 'Juan LERENA (m. Eduviges Pelaez 1894) is known from ONE LINE of the Montevideo dispensation register and nothing else - with no age, no birth and no parents.' The register carried two rows.

now THEY ARE ONE MAN. The baptism of 26 November 1905 at Santisima Trinidad names the parents the dispensation ledger withheld: 'hijo legitimo de Juan Lerena y Eduviges Pelaez. Abuelos paternos: BERNABE y CASIMIRA PELAEZ'. Juan Bernardo Lerena Pelaez, born 20 August 1871, is the only Juan among the children of Bernabe Lerena and Casimira Pelaez. HE WAS TWENTY-TWO WHEN HE WAS DISPENSED. The rows are merged.

how Searching the full text for the BRIDE rather than the groom, then looking for a CHILD rather than a marriage - the same method that answered the 1913 couple an hour earlier. THE SPLIT WAS NOT BAD REASONING; THE EVIDENCE ARRIVED A DAY LATE. What it shows is that 'no parents are recorded' is a statement about a ledger, never about a man.

was That the Arquidiocesis de Montevideo MARRIAGE DISPENSATION RECORDS are the Vicar General's SUMMARY LEDGER, one sentence per couple, naming no parents - and that the INFORMACIONES MATRIMONIALES are 'a separate series' this archive had not reached. Published 8-9 October on /uruguay/ and in data/montevideo-dispensations-fulltext.tsv, itself a correction of an earlier claim that the dispensations DO name parents.

now THE COLLECTION CONTAINS BOTH, AND THE DIFFERENCE IS THE DECADE, NOT THE SERIES. The volumes read on 8 October were all 1885-1913 and are indeed a summary ledger. THE VOLUMES OF THE 1850s IN THE SAME COLLECTION ARE FULL EXPEDIENTES: petition, numbered interrogatory, sworn depositions naming parents AND grandparents of both parties, the Cura Vicario's report - and, on the Bernabe Lerena expediente of 15 February 1858, A KINSHIP DIAGRAM THE PRIEST DREW HIMSELF. Three were read on 9 October (1854, 1855, 1858) and all three name four generations.

how By doing what the ledger finding said was pointless: searching the dispensation collection for a woman whose parents were wanted. THE ORIGINAL CLAIM WAS TRUE OF EVERY VOLUME THIS ARCHIVE HAD OPENED AND FALSE AS A STATEMENT ABOUT THE COLLECTION - a generalisation from one date range, published as a fact about a record class. IT IS THE SAME SHAPE AS THE SOUTH AFRICA ERROR OF THE DAY BEFORE: a true observation about what was tried, written down as a conclusion about what exists.

was That the 1837 San Jose marriage act and the 1858 Trinidad expediente give the SAME PLACE OF ORIGIN - published hours earlier in data/trinidad-founding-kindred.tsv, on /uruguay/ and in the register row for Ramon LERENA of Rosario-Colla, as corroboration for identifying the Colla Ramon with the Ramon of the Trinidad kinship diagram: 'same names, same generation, same place of origin'.

now THEY GIVE DIFFERENT BIRTHPLACES. The 1837 act, read off the plate, says Mauricio Lerena was 'natural de la VILLA DEL COLLA'. The 1858 expediente says the Trinidad Lerenas are 'naturales de SAN JOSE'. WHAT IS ACTUALLY SHARED IS NARROWER: Mauricio was MARRIED at San Jose de Mayo and they were BORN there. The identification loses one of the two legs it was standing on.

how BY DOING THE TEST THE CLAIM ITSELF NAMED. The row said the weakness was that the 1837 act was 'INDEXED, not image-read' and set reading the plate as the test. The plate was read the same day and corrected the paragraph instead of confirming it. THE ERROR WAS WRITING A CONCLUSION FROM AN INDEX FIELD ('Birthplace: Colla') WITHOUT NOTICING IT CONTRADICTED THE SENTENCE BEING BUILT - the index said Colla all along, on the very page that was being cited.

was That Cipriano LERENA of Trinidad and Cipriano LERENA Montana were two people. Carried as separate register rows since 21 September; the second row itself noted the forename 'is also one of the Trinidad men' and held them apart at index level.

now ONE MAN. The Flores birth act of his daughter Carmen Donata, 2 March 1895 at El Tala, has him declare in person at forty-two, names his wife LAUREANA BELOQUI and his parents MAURICIO [LERENA] and CASIMIRA MONTANO. THAT IS THE SPOUSE AND BOTH PARENTS IN ONE ACT. The rows are merged, and three children already in this register become grandchildren of Mauricio Lerena of Colla.

how Searching the full text for the grandmother's name, CASIMIRA MONTANO, rather than for either Cipriano. THE PAIR WAS CORRECTLY FLAGGED AND CORRECTLY NOT MERGED FOR EIGHTEEN DAYS, which is the system working: it was waiting for an act, and an act came.

was That the identification of the Trinidad kindred's RAMON LERENA with the RAMON LERENA of Rosario-Colla was an OPEN CANDIDATE, 'NOT MERGED', resting on a name match. Published twice on 9 October - in data/trinidad-founding-kindred.tsv, data/colla-trinidad-bridge.tsv, two register rows and two sections of /uruguay/ - with a test proposed and a caution that 'two couples can share two names'.

now THIS ARCHIVE HAD ALREADY MADE THE IDENTIFICATION ON 17 SEPTEMBER, AND SAID SO IN ITS OWN FILES. data/ramon-lerena-rosario.tsv states outright: 'Bernabe is Alejandro's son; Cipriano is Mauricio's; Alejandro and Mauricio are brothers', draws the whole descent from Ramon Lerena and Maria Josefa Gonzalez, and names six of Alejandro's children including BERNABE and SOCORRO. data/descent-spines.tsv carries Ramon's house as a graded Documented spine, and /uruguay/ has been drawing it for three weeks.

how BY SEARCHING THE INDEX FOR 'ANY RECORD NAMING RAMON WITH MORE THAN ONE CHILD' - the test this archive set itself - AND FINDING ITS OWN ANSWER FILE. THE FAILURE WAS NOT READING THIS ARCHIVE'S OWN DATA BEFORE PUBLISHING A WEAKER CLAIM THAN IT ALREADY HELD. IT IS THE THIRD INSTANCE OF ONE PATTERN - the 1891 Northam census and the FreeBMD death sweep were both reported as undone when they were done. EACH TIME THE CLAIM WAS WRITTEN FROM MEMORY WHILE FINISHING SOMETHING ELSE, AND EACH TIME A FILE IN data/ ALREADY HELD THE ANSWER.

was That BERNABE LERENA of Trinidad was unattached - 'Nothing joins him to these two, and the forename is not proof' (register row), and that a third Bernabe 'sits in this register unattached'. Carried in the register since 20 September.

now THE REGISTER WAS CONTRADICTING THE ARCHIVE'S OWN EVIDENCE FILES FOR THREE WEEKS. data/ramon-lerena-rosario.tsv had him as Alejandro's son since 17 September and /uruguay/ drew it; the register said unattached. BOTH ARE NOW SUPERSEDED BY DOCUMENTS READ ON 9 OCTOBER - his own marriage expediente of 15 February 1858, which names his parents and carries the priest's kinship diagram, and his father's marriage act of 16 September 1825, read off the plate.

how BY THE REGISTER AND THE EVIDENCE FILES BEING COMPARED FOR THE FIRST TIME. This is the defect the owner's question - 'is all this research making it to our sites, persons, families charts, diagrams?' - was pointing at, asked twice. THE ANSWER WAS NO IN A DIRECTION NOBODY CHECKED: not the register falling behind the site, but THE REGISTER FALLING BEHIND data/, with the site rendering the newer of the two and the register rendering the older, both on the same deploy.

was That lot 210 of Nuno Fernando Lerena's deeds folio lies in 'LONDON STREET'. Published hours earlier in data/south-africa-fulltext-closed.tsv, on /south-africa/, in his register row and in work-list row 197 - read off the plate at a magnification that was not enough, and then built on.

now THE PLATE READS 'LENDIN', NOT 'LONDON'. At full magnification the second letter is an unmistakable e, not an o, and there is an n before the d where London has none; the i-dot sits after the d. THE STREET IS NOT CONFIDENTLY READ and is now recorded as 'Lendin (so written; not resolved)'. IT IS NOT DEMONSTRABLY 'LEIDEN' EITHER, although that is the likeliest intended name - Nuno lived at 24 Leiden Road - because Leiden would put the i before the d and this hand does not.

how BY CHECKING THE STREET AGAINST THE PLACE INSTEAD OF ACCEPTING IT. Gerdview's streets are named for Dutch and German cities - Leiden, Weert, Cologne - and 'London' did not fit the scheme, which sent the plate back for a harder look. THE LESSON IS THE MAGNIFICATION: the first read was done at a zoom that resolved 'Lerena' and 'Nuno' perfectly and was still too coarse for a six-letter word in the same hand. A READING IS ONLY AS GOOD AS THE SMALLEST THING IT RESTS ON.

was That the Lerena cards of army drawer CAJON 985 were 'sitting unread', and that the drawer is 'filed alphabetically'. Both published on /open-questions/ under question 1, the question this archive exists to answer, since 10 September 2026 - a page whose stated purpose is 'published so strangers can help'.

now THE DRAWER WAS READ ON 10 SEPTEMBER, THE SAME DAY THE QUESTION WAS WRITTEN, AND THE FATHERS TRACED ON THE 13TH. data/cajon-985-scan.tsv holds all 39 frames, images 650-688, with every card's indexed parents; data/cajon-985-origins.tsv traces the fathers to three Uruguayan branches. AND IT IS NOT ALPHABETICAL: the same file records that the drawer is GROUPED BY SOUND - Lera, Lerena, Lorena, Serena, Sera, Era, Leza, Perena, Perella and Hereijo interfiled - which is a finding in its own right and the reason the index's manglings survived in it. TWENTY LERENA CARDS, AND NO JUAN CARLOS ON ANY OF THEM.

how BY GOING TO DO THE WORK THE PAGE ASKED FOR AND FINDING IT DONE. THIS IS THE WORST PLACE IN THE ARCHIVE FOR IT TO HAVE HAPPENED: /open-questions/ exists so that a stranger can help, and its first question was sending them to a drawer that had been read, by a filing rule that is wrong. IT IS THE SAME DEFECT AS THE REGISTER CALLING BERNABE UNATTACHED - a page that stopped where data/ went on - and the fifth instance recorded today.

was That the four untraced fathers of Cajon 985 include 'the RAMON of the Concordia pair'. Written into /open-questions/ question 1 an hour earlier, when its ask was rewritten.

now THE CONCORDIA RAMON IS TRACED. data/cajon-985-origins.tsv has NEMESIO RAMON LERENA married at SALTO, URUGUAY, 16 January 1903, then at Concordia, Entre Rios, directly across the river - one of the three branches the same paragraph calls traced. THE FOURTH UNTRACED FATHER IS ARTURO LERENA, husband of Amalia Diaz, with sons born in La Pampa, Mendoza and Entre Rios 1889-1897.

how BY GOING TO SEARCH THE FOUR AND FINDING ONE OF THEM ALREADY ANSWERED. The same move that caught the drawer, one level down. AND ARTURO IS THE ONE OF THE FOUR THAT THIS ARCHIVE HAS ALREADY GOT WRONG ONCE: his row was first written as 'Montevideo 1858, son of Avelino Lerena and Ambrosia Traibel' and withdrawn the same day as a conflation with JUAN ARTURO LERENA TRAIBEL, whose wife was Ema Regules. A NAME THIS ARCHIVE HAS ALREADY MISIDENTIFIED ONCE DESERVES THE MOST CARE, NOT THE LEAST.

was That the unmarked LERENA grave holding three people is at NEWLANDS - recorded on work-list row 106 from the cemetery office via Cheryl's sister Patricia, and carried since September.

now TWO OF THE THREE ARE AT MAITLAND ROAD CEMETERY, IN GRAVE 7719. The Record of Interments, page 4706 entry 423, puts CAREEN MARIA LERENA, aged two weeks, in Roman Catholic allotment, Cons. No. 1, grave 7719. This archive already held MARY SEPTIMA TAYLOR at "Maitland Road Cemetery interment register p.3215 entry 163, Nov 1935 ... Cemetery no 1, grave 7719A". Same cemetery, same allotment, same number. The likeliest source of the error is that PABLO ARMANDO DIED at Newlands, which is a different Cape Town suburb from Maitland. AND LATER THE SAME DAY IT WAS CORRECTED THE REST OF THE WAY. This row first said 'what is not corrected: whether he is in that grave. He died in 1950 and this opening is 1947, so this page cannot say.' The 1950 volume was then found and read. HE IS IN IT. Maitland Road Cemetery, Record of Interments, page 339 entry 327: LERENA, Robert Paul, M.E., 76, of Newlands, 7 Forth Road, Roman Catholic, Cemetery No. 1, grave 7719 C, 8 feet, Father Plotton, undertaker Goodall & Wms - and in the last column, 'In Plot of R. P. Lerina'. SO THE FAMILY WERE RIGHT ABOUT EVERYTHING EXCEPT THE CEMETERY. Three people, one grave, unmarked but for the surname: A, B and C of plot 7719, which was his own. They were told Newlands and it is Maitland.

how BY DAVID SENDING THE REGISTER PAGE. The archive had the family's account of the grave and its own citation for Mary Septima's burial, and had never put the two side by side - the grave number it already held, 7719A, was sitting in the register beside a cemetery name that contradicted the story.

was That Careen Maria's grave was 7719, and that the resolution could not rule out an 'A' suffix. Published this morning, from the page David supplied.

now IT IS 7719 B. Read at full resolution the same afternoon - 4,863 pixels across the opening against about 1,100 - the suffix is plainly a B, and the plot has three lettered places: 7719 A Mary Septima, 1935; 7719 B Careen Maria, 1947; 7719 C Robert Paul, 1950. THE MORNING'S READING WAS NOT WRONG SO MUCH AS BLIND: the digits were right, the suffix was unreadable, and this archive said so at the time rather than guessing. The guess on offer was 'A', to match Mary Septima - and 'A' is the one letter it could not have been.

how BY GOING TO THE SOURCE INSTEAD OF THE PHOTOGRAPH OF IT. The supplied scan carried the black bars that FamilySearch's Cape cemetery collection puts on its own images, which is what identified the route; signing in gave the same plate at four times the linear resolution.

was That MAURICIO LERENA of Colla, born about 1815, died on 3 March 1925. Carried in data/last-attested-alive.tsv as an explicit death date on this person's row, tier DEATH - the tool's most confident class.

now IT IS HIS DAUGHTER'S DEATH DATE. 3 March 1925 belongs to Clara Lerena de Pallares (b.1848), named in his own note along with Ignacia Serena Leren de Dolz, d. 12 March 1921. He was born about 1815, so the archive was asserting he died at about 110, and nothing flagged it. HE HAS NO ATTESTED DEATH DATE AT ALL, and now says so: BIRTH ONLY, a floor and nothing more.

how BY A TOOL BEING FIXED FOR A DIFFERENT REASON. tools/last-alive.py took the FIRST 'died <date>' anywhere in a row's prose. That broke on 8 October when Pablo Armando's note gained a sentence naming the three burials in his own grave, and his death jumped to the infant's 1947 - which check-spines caught, because the /uruguay/ ladder still said 1950. The repair was to corroborate death dates against an independent record year instead of ranking them by position, and the moment it ran it reported Mauricio as unresolvable. THE BUG BIT HARDEST WHERE THE RESEARCH WAS BEST: the fuller a note, the likelier it names somebody else's death, and the tool silently preferred whichever came first in the string.

was That the Taylor family group photograph shows 'at least eight daughters and two sons', and that the number is the ordinary explanation of the name SEPTIMA, meaning seventh. Published this morning, from the version David sent in conversation.

now YOU CANNOT COUNT DAUGHTERS IN A PHOTOGRAPH. The figure count was also wrong - at full resolution there are twelve people and a dog: three male, eight female, and one infant whose sex cannot be told - but the figure count was never the error that mattered. A GRANDDAUGHTER, A NIECE, A DAUGHTER-IN-LAW AND A NEIGHBOUR'S CHILD ALL PHOTOGRAPH IDENTICALLY. There are two seated adult women in the group and nothing in the image says which is the mother of the small girl; the one in the bonnet and fringed shawl reads as the elder, which is as consistent with a grandmother as with a mother of ten. THE PAGE NOW COUNTS FIGURES AND STOPS THERE, and names the record that would actually settle the sibling count: the 1891 census of Northam, already read here for the Brayleys and never counted for the Taylors.

how BY DAVID, IN ONE SENTENCE, ON READING THE PUBLISHED PAGE. 'I think there is a grandmother or aunt in that photo as well.' THIS ARCHIVE'S OWN STANDARD SHOULD HAVE CAUGHT IT FIRST: it says on /south-africa/ that it 'does not turn consistent with into is', and then did exactly that with a picture - counted faces, assumed every one of them was a child of the couple, and ran the total back into a name it already believed.

was That "Mary Septima stayed with her sister EDIE at 26 Fore Street, Northam, in 1935" is independent evidence for a missing sister named EDITH. Carried since 16 September in data/brayley-bmd.tsv, data/1891-northam-census.tsv and data/brayley-two-questions.tsv, where it is called knowing the name "from a completely different direction".

now EDIE IS ISABELLA. This archive records her in three other files as Isabella Temple "Edie" Taylor, who married HENRY DALLING at Northam on 10 September 1900. data/family-memory.tsv names the hostess of 1935 as "her sister Edie DALLING" - the married surname is in the sentence. data/passenger-lists.tsv gives Mary Septima's 1935 address in Britain as "C/O DALLING, FORE ST, NORTHAM" and says in terms it is "her elder sister Isabella Temple 'Edie' Taylor". And the 1939 Register for 26 Fore Street lists ISABELLA T DALLING, born Taylor, 10 September 1874. THE EDITH ARGUMENT WAS COUNTING ONE WOMAN TWICE.

how BY DAVID ASKING FOR THE 1891 NORTHAM CENSUS TO BE DONE FOR THE TAYLORS. It had been done three weeks earlier; what the request produced instead was a re-reading of the file, and the "Edie" sentence did not survive it. THE CANDIDATE ITSELF IS NOT DEAD - the tree still says ten children and names nine, Septima still means seventh, and the birth index still offers one Edith in the gap - but it has lost a leg it should never have had.

was That the 1891 census of Northam was "already read here for the Brayleys, and never counted for the Taylors". Published this morning on /family-papers/ as the record that would settle the sibling count.

now IT WAS READ FOR THE TAYLORS, ON 16 SEPTEMBER 2026, and it is the Taylor household that the file is named for. data/1891-northam-census.tsv opens with the heading "THE HOUSEHOLD AT 2 FORE STREET, NORTHAM" and gives ten people by name, age, occupation and birthplace. The Brayleys were the SECOND half of that evening's work, not the first. AND IT WOULD NOT HAVE SETTLED THE COUNT ANYWAY: Mary Septima is ABSENT from it, and so is any Edith of her age.

how BY DOING WHAT WAS ASKED AND FINDING IT ALREADY DONE. The claim was written from memory while correcting a different error, which is exactly when this archive has gone wrong before - reaching for a record name to end a paragraph with instead of opening the file.

was That a seventh daughter, EDITH, born at Bideford in September quarter 1882, fills the gap the name SEPTIMA requires. Graded a STRONG CANDIDATE since 16 September, on three legs: the tree's count of ten children, the meaning of Septima, and a single birth-index entry.

now SHE IS NOT IN THE FAMILY'S BAPTISM REGISTER, and every one of her supposed siblings is. Devon Baptisms, Northam St Margaret, archive reference 1843A/PR/1/16: SEVEN children of William Taylor and Mary Ann Helen/Ellen - Horace Archibald and Isabella Temple (1874), Margaret Helen Walker (1876), Irene (1879), Lucy Elizabeth (31 Aug 1881), Mary Septima (30 Oct 1884) and Agnes Walker (21 Aug 1889). NO EDITH. Lucy and Mary Septima sit three years and two months apart with nothing between them, and the candidate falls exactly in that gap. AND THE ONLY TWO DEVON EDITHS NEAR THE DATE ARE OTHER PEOPLE: Edith Taylor, baptised 21 Nov 1883 at Bideford, mother MARIA; Edith Mary Taylor, 8 Apr 1885 at Bideford, mother ELIZABETH. THE CANDIDATE IS DOWNGRADED FROM STRONG TO WEAK, and what survives is only the tree's unsourced count and the meaning of a name.

how BY SEARCHING ON THE MOTHER INSTEAD OF THE CHILD, after David asked for the 1891 census. One search on 'Mary Ann Helen' returns the whole sibling group at once, which is what makes the absence provable - child-by-child searching could only ever have failed to find her. THE CONTROL WAS RUN FIRST: Agnes Walker Taylor, whose birth this archive had to the day and the address, was looked for before anything else and is there. AND IT STRENGTHENS DAVID'S OWN READING OF THE PHOTOGRAPH - if there is no eighth daughter, the eighth woman in the group is not a daughter, and his grandmother or aunt is the better explanation.

was That William Taylor and Mary Ann Hellen Brayley had TEN children, and that a tenth is missing. Taken from a MyHeritage tree that names nine and asserts ten, and carried on /taylor/ under a Documented chip. It has driven three weeks of searching: work-list rows 126, 127, 128, 192 and 193, a FreeBMD death sweep, a FreeREG attempt, a findmypast baptism search and two candidate children.

now THERE ARE NINE, AND NOTHING IS MISSING. The GRO's own index, which prints the mother's maiden surname, was searched for TAYLOR births with mother BRAYLEY across 1870-1893, both sexes, ten windows: Kathleen Georgina (1870 D, 05B/490), William John (1870 D, 05B/489), Horace Archibald (1872 J, 05B/507), Isabella Temple (1874 D, 05B/497), Margaret Hellen Walker (1876 D, 05B/481), Irene (1879 M, 05B/524), Lucy Elizabeth (1881 S, 05B/508), MARY SEPTIMA (1884 D, 05B/506) and Agnes Walker (1889 S, 05B/502). All at BIDEFORD. Nothing before, nothing between, nothing after.

how BY DAVID LOGGING THIS SESSION IN TO gro.gov.uk and the search row 128 had named in September being run at last. THE INDEX ANSWERS BY DEFINITION RATHER THAN BY INFERENCE: FreeBMD's birth index carries no mother's maiden name, so no amount of work there could ever have joined a Bideford Taylor child to this household - which row 128 said in terms and was right about. AND THE CRITICAL WINDOW WAS RE-RUN ON LOOSE SPELLING MATCHING because this archive has documented BRAILEY as a variant: 28 births across England, the same two at Bideford.

was That EDITH Taylor (Bideford, Sep q 1882) and ETHEL Taylor (Bideford, Dec q 1882) were candidates for the missing daughter, and that an unnamed TWIN OF IRENE, 1879, was the fifth daughter. Edith was graded a STRONG CANDIDATE on 16 September; Ethel was called the better fit on 17 September; the twin has been on /taylor/ since the beginning, inferred to be female 'or the arithmetic does not come out'.

now ALL THREE ARE GONE. Neither Edith nor Ethel has BRAYLEY as her mother in the GRO index, so neither is a child of this couple. And IRENE IS ALONE IN 1879: no second birth to this mother that year, of either sex. The twin was never a person. She was an arithmetical requirement wearing a tree's silence as evidence, and this archive built her, published her, and reasoned from her.

how BY ASKING THE ONE INDEX THAT CAN REFUSE AN ANSWER. Every earlier search could only fail to find a child; this one can show she is not there, because it is keyed on the mother. THE BAPTISM REGISTER HAD ALREADY POINTED THIS WAY hours earlier - seven children at Northam, no Edith, no Ethel, Irene alone in 1879 - and the GRO settled it.

was That SEPTIMA means seventh and that the name is therefore a count of daughters - 'her own name is a piece of arithmetic, and it turns out to be exactly right'. The founding observation of /taylor/, and the reason the tenth child was ever looked for.

now MARY SEPTIMA IS THE SIXTH DAUGHTER AND THE EIGHTH CHILD, and no reading of the registered family makes her seventh of anything. Daughters in order: Kathleen (1), Isabella (2), Margaret (3), Irene (4), Lucy (5), Mary Septima (6), Agnes (7). THE NAME IS NOW AN UNEXPLAINED FACT RATHER THAN AN ARITHMETICAL ONE, which is a worse answer and a truer one. ONE CATEGORY OF CHILD LEAVES NO BIRTH REGISTRATION AT ALL: a stillbirth, which was not registrable in England and Wales until 1927. A seventh pregnancy carried to term and lost would be invisible to every index this archive can reach, and would still be counted by the woman who bore her. THAT IS A POSSIBILITY, NOT A FINDING, and it is recorded as the shape of the remaining doubt rather than as its answer.

how BY FINISHING THE COUNT INSTEAD OF DEFENDING IT. The arithmetic was never checked against a complete list of children, because no complete list existed until today.

was That John Walker Brayley was born at ST PANCRAS (the 1851 census) or at ST PAUL'S (the tree), and that his birth year was unresolved between the tree's 1845 and a death index implying 1847. data/1851-bideford-brayley.tsv left the parish disagreement open in terms.

now HIS BIRTH IS REGISTERED IN THE HOLBORN UNION, June quarter 1845, Volume 02 page 149 - indexed as BRAILEY, JOHN WALKER, mother TEMPLE. THE YEAR IS SETTLED FOR THE TREE: a birth on 22 June 1845 falls in the June quarter, so the 1900 death index that made him 53 had a careless informant - which this archive had already concluded from the census. THE PARISH DISPUTE RESOLVES TO NEITHER SIDE: the district is HOLBORN, which is also where his sister ISABELLA was born - the 1861 census enters her as 'London (City), ST ANDREW', that is St Andrew Holborn. AND HIS MIDDLE NAME IS WALKER, which pushes this family's naming discipline a full generation earlier than the archive had documented it: Temple and Walker were being used as forenames in 1845, not first in 1874.

how BY SEARCHING FOR THE MOTHER INSTEAD OF THE CHILD, in the GRO's own index. The search was run to settle a different question entirely - whether Mary Ann Hellen Brayley has a birth registration - and he came back with it.

was That MARY ANN HELLEN BRAYLEY's birth registration is merely NOT YET FOUND. data/brayley-bmd.tsv recorded on 14 September that FreeBMD holds no Mary Brayley and no Mary Brailey in the June quarter of 1843 anywhere in England and Wales, with the control passed - and left it as an absence awaiting a better search.

now SHE WAS ALMOST CERTAINLY NEVER REGISTERED, and the archive should stop looking. The GRO's own index was searched on her MOTHER's maiden surname, TEMPLE, with the child's surname on loose matching, across 1838-1852, both sexes. It returns exactly two children of this couple: JOHN WALKER (1845, Holborn) and GEORGINA MARGARET (1850, Bideford). THE TWO EARLIEST CHILDREN ARE BOTH ABSENT - Isabella, about 1841, and Mary Ann Hellen, 20 May 1843, both born in London - and the two later ones are both present. Birth registration was not compulsory on parents until the Act of 1874, in force from 1875; under-registration in the 1840s is well attested among the urban poor; and the 1851 census describes this household as 'Shoemaker Wife Pauper'. The null is now double and arrives from two directions - FreeBMD under her own name, the GRO under her mother's.

how BY ASKING A DIFFERENT QUESTION OF A DIFFERENT INDEX. Searching for a person who may not be in a book can only ever fail; searching for her PARENTS returns the children who ARE there, and the shape of who is missing. AND THE SPELLING CONTROL IS BUILT IN: the query said BRAYLEY and the index returned BRAILEY twice, so the nulls are not a spelling artefact.

was That ALEJANDRO LERENA (m. Elisa Villardemoros) and ALEJANDRO M. LERENA (with Julia Bassani) are "DIFFERENT MEN, on wives", because "the other was a WIDOWER living apart from JULIA BASSANI at calle San Jose N 17 in 1886", and that "no record joins them". Written by this archive into data/_person-pairs.txt on 22 September 2026.

now THE ARGUMENT WAS BACKWARDS AND THE FACT WAS WRONG. Being a WIDOWER in 1886 does not separate a man from a wife he had in 1872 - it is precisely what a first marriage leaves behind. The strongest link in the evidence was written down as the objection to it. AND A RECORD DOES JOIN THEM: the Iglesia Matriz christening of 29 September 1872, read off film 004298263 image 03147 on 27 September 2026, gives the godparents as "ALEJANDRO M. LERENA y ELISA VILLADEMOROS". The M. is carried by no other Alejandro in this register. THEY ARE STILL NOT MERGED - an initial and a wife is not proof, and this archive has merged on less and been wrong - but the pair declaration now says the right thing, and names the two records that would settle it.

how BY OPENING A PLATE THE INDEX HAD DATED 1919. The record surfaced in a search for a Montevideo marriage and looked like a 1919 christening with Alejandro M. Lerena as a godfather. The act is 1872; the 1919 is the certificate. Had the index date been believed, the record would have been discarded as a man too old to be ours.

was That ALEJANDRO MAXIMO LERENA was "Born 1839 Uruguay, WHICH THE ARGENTINE RECORD INDEPENDENTLY CONFIRMS" - stated in data/montevideo-lerena-trunk.tsv, graded "PRIMARY (indexed)" in data/cajon-985-origins.tsv against the source "FamilySearch: Argentina, Capital Federal, Catholic Church Records 1640-1978", used in data/the-three-alejandros.tsv to hold two men six years apart, and published on /uruguay/ in the descent spine as "Alejandro Maximo (1839-1903)".

now NO ARGENTINE RECORD CONFIRMS IT, BECAUSE NONE OF THEM STATES HIS AGE. All four Capital Federal records for this man were opened on 22 September 2026 - the baptisms of Elisa Cornelia, Carmen Romualda and Maria Luisa, and the marriage of his son Hector - and EVERY ONE OF THEM LEAVES THE FATHER'S AGE BLANK, while giving the MOTHER's (Julia Bassani, 40 years in 1905 and 39 in 1907). The 1839 is a FAMILY TREE VALUE carried from PID KZ2N-84J and re-graded as a record on the way in. AND IT IS FIVE YEARS OUT. Two Montevideo birth acts of his own children, read off the plate the same day, take his age from his own mouth: "de cuarenta y dos anos de edad" on 6 November 1886 and "de 43 anos de edad" on 16 March 1888, which puts his birth between 17 March and 6 November 1844.

how BY ASKING WHERE A NUMBER CAME FROM. The tier said PRIMARY (indexed), so the index was opened to see the entry - and the field was empty. A tier is a claim about a source and it can be checked by opening the source, which is the only thing that ever checks one.

was That AVELINO LERENA FERNANDEZ died on "23 August 1890" at "La Paz, Canelones", and that he was "the seventh of eight children" - both printed in the descent spine on /uruguay/, the first of them since the page was built.

now HE DIED ON 22 AUGUST 1890 AT MONTEVIDEO, AND WAS REGISTERED THE NEXT DAY. That is what his own register row says, and data/last-attested-alive.tsv, data/1852-aguada-union-certificates.tsv and data/alberto-lerena-larriera.tsv all say the 22nd. THE 23rd IS THE REGISTRATION DATE and "La Paz, Canelones" appears nowhere in this archive outside one row of data/montevideo-lerena-trunk.tsv - a FamilySearch tree file whose own header reads "TIER: SECONDARY THROUGHOUT ... no row here may be cited as fact". The page cited it as fact. AND HE IS THE EIGHTH OF EIGHT CHILDREN, not the seventh - the reconciled list runs 1793, 1795, 1799, 1800, 1804, 1806, 1811, 1815 - or the seventh of seven if FIDEL, the only one of them attested at second hand, is set aside. There is no count under which "seventh of eight" was right.

how BY AUDITING THE HAND-WRITTEN LADDERS AFTER A SECOND ONE FAILED THE SAME DAY. Gen 4 of this same spine had just been corrected for asserting "Alejandro Maximo (1839-1903)". tools/check-inline.py had been reporting all three ladders for days as "reachable only by editing the page" and nothing gated them. ALL THREE ARE NOW GENERATED FROM data/descent-spines.tsv, so the checks that read data can reach them.

was That 'Juan Carlos Lerena as a named person, any role, any country, any collection' returns ONE record and it is not him, and that there is 'no indexed child of ANY man recorded exactly as Juan Carlos Lerena'. Published on /direct-line/ since 17 September, in the table that carries this archive's central null.

now BOTH RETURN TWO NOW, AND THE SECOND IS HIM. Re-run 21 September while testing the exact-forename trap: "Juan Carlos Lerena, Father, Argentina Military Records 1911-1936, spouse MARIA LERENA, child PABLO ARMANDO LERENA" - ark 1:1:6TH8-DVXH, which is the 1908 enrolment card and a record THIS ARCHIVE HAS CITED SINCE IT FOUND IT. NOTHING WAS WRONG WITH THE SEARCHES. The Argentina Military Records index has since gained the card; its FamilySearch citation reads 16 September 2026, the day after the test. WHAT DOES NOT CHANGE: Juan Carlos still has no record of his own - no birth, no marriage, no death - and the new hit is his son's record naming him as a father. WHAT DOES: he is findable by name now, and the index places him at "Rosario de Lerma, SALTA", the error this archive corrected on 13 September.

how Re-running the central null in the fuzzy-forename/exact-surname shape, which is work list row 157 - the trap sweep found the archive's own document instead

was That ZELMIRA CARLOTA LERENA LARRIERA, born in the Departamento de San Jose on 4 November 1880, 'is NOT ours' - published in her register row since 14 September 2026, which added that her father Alberto Lerena 'would be the right age to be ALBERTO OCTAVIO LERENA LENGUAS' and that 'NOTHING JOINS THEM but a forename and an age'.

now SHE IS OURS, and the candidate was the wrong man. Alberto Lerena's own Montevideo death registration of 2 March 1921, certificate 246, names his wife MARIA INES LARRIERA and his parents AVELINO LERENA and AMBROSIA TREIBEL. So Zelmira's father is PEDRO ALBERTO YSABELINO LERENA TRAIBEL, baptised at the Aguada on 21 November 1852, and she is a GREAT-GRANDDAUGHTER OF SANCHO - through Avelino, not through Luis. The row was right that the question turned on her father and right to refuse the guess; it was wrong to write off the person while the question was open. Seven brothers and sisters of hers were entered the same day. THE LINK WAS INDEX-LEVEL FOR ABOUT AN HOUR. The 1921 image was opened the same day - film 008165205 image 01604 - and reads 'Que el finado era hijo de AVELINO LERENA, fallecido, y de AMBROSIA TREIBEL, fallecida', and then counts NUEVE HIJOS and names them. So the retraction above stands on an image, and the act supplied a NINTH child the index never showed. See data/alberto-lerena-larriera.tsv.

was That Alfredo Justiniano Lerena Traibel married "HERMENEGILDA CAROLINA GAVARRO" , daughter of "Don GERONIMO GAVARRO" and "Dona CAROLINA JUAREZ" - three register rows, a dispensation table, the 1892 working log and the last-attested-alive file, published since 14 September 2026.

now ALL THREE SURNAMES ARE WRONG, AND THE FAULT IS THIS ARCHIVE'S OWN RULE, UNKEPT. The 1892 marriage act was recorded as FULL-TEXT TRANSCRIPT READ - a MACHINE transcript - and the image was never opened, which is the one thing this archive says it always does. The image was opened on 21 September 2026: film 007713383, image 02601, folio 340. It reads 'hija legitima de Don GERONIMO GAVAZZO y Dona CAROLINA SUAREZ'. So she is a GAVAZZO and her mother a SUAREZ. It is corroborated by a second image forty-eight years later - her own Montevideo death registration of 20 June 1940, which writes her HERMENEGILDA GAVAZZO DE LERENA with the same two parents. Two images against one machine transcript. GAVAZZO is a known Montevideo surname and GAVARRO is not, which is exactly the kind of tell that should have prompted the check earlier.

was That MARIA FERNANDEZ TRESGUERRAS, Sancho's wife, was born at "Montevideo" - the birthplace this register carried for her, and the place the atlas put her.

now SHE WAS NATURAL DE ASTURIAS, and this archive had known it for a week without changing the row. The 1811 baptism of her son Luis Escolastico reads 'y de D.a MARIA [Fernandez] nat.l de ASTURIAS'. data/1811-luis-baptism-act.tsv recorded that on 14 September 2026; the register was never updated, so the correction existed and never propagated - the exact fault this archive has a standing rule about. IT ALSO CHANGES THE FAMILY'S SHAPE: both halves of Sancho's household were SPANISH IMMIGRANTS, from opposite ends of Spain - Andalusia and Asturias - meeting at Montevideo. And her parents RODRIGO FERNANDEZ and ROSA FERNANDEZ, named on the same act, had never been entered as people either. Read off the plate 21 September 2026: film 007713375 image 02410.

was That the two Moratorio Lerena baptisms of 8 March 1880 took place at "UNION, MONTEVIDEO" , published across six register rows, the working log, and /uruguay/ - where it had grown a conclusion: that Avelino Pedro Joaquin Silvestre was christened at the Union in 1866 "the same parish where his cousins the Moratorio girls were baptised fourteen years later" .

now THEY WERE BAPTISED IN THE CATEDRAL BASILICA DE MONTEVIDEO. Read off film 007713378 image 02598 on 21 September 2026: 'con licencia del infrascripto Cura Parroco DE ESTA BASILICA MENOR Y CATEDRAL DE MONTEVIDEO, bautice solemnemente EN ELLA a Maria Carlota'. The UNION came from the INDEX'S PLACE FIELD, which this archive copied as though it were testimony. The cousins were not baptised in the same parish, and the sentence that said so was built on a place nobody had read. The same plate settled something else in the archive's favour: both acts state 'que nacio el VEINTE Y CUATRO DE FEBRERO ultimo', so the twins reading flagged as an untested inference is CONFIRMED.

was That Bernabe Lerena married "FLORENCIA CARDOZO" - published across five register rows and the 1911 dispensation working log since 14 September 2026.

now SHE IS FLORINDA. The dispensation folio is not prose but two kinship diagrams the priest drew himself, and her name is written in both: 'Bernabe Lerena casose con FLORINDA CARDOZO' and 'FLORINDA CARDOZO casada con Bernabe Lerena'. Read off film 007713362 image 02473 on 21 September 2026. The same plate gave two people the transcript never carried into the register - BERNABE LERENA THE ELDER and CUSTODIO CARDOZO, the apex of each schema, and the generation above the 1911 couple.

was That Victor Lerena's mother was "JOSEFA SEGUERA" , and that in his 1889 death registration the surname is "written LORENA for him and LERENA for his son and his father in the same sentence" .

now SEGUEIRA, and the sentence says LERENA throughout. Read off film 008165116 image 01442: 'el finado era hijo de PEDRO LERENA y JOSEFA SEGUEIRA, orientales, finados. Abuelos [paternos] FIDEL LERENA y MARIA RUEDAS: maternos JOSE SEGUEIRA y RAMONA...'. LORENA is the MARGIN HEADING - 'Num. 622 Victor Lorena' - which is the clerk's index entry and not the text. This archive read a margin as though it were the sentence, and built a claim about the family's spelling on it. The confusion of L and Ll in this family is real; this instance of it was not. The plate also gave a cause of death the transcript never carried - TUBERCULOSIS PULMONAR - and two people nobody had entered, JOSE SEGUEIRA and RAMONA.

was That the 1833 godfather is rendered "Abelino Serena" , and that this is "one more instance of the S-for-L this archive has met before" .

now THE PLATE SAYS LERENA. Film 007713376 images 01176-01177, read 21 September 2026: 'Padrinos ABELINO LERENA y DOMINGA LERENA'. The S-for-L was the MACHINE'S, not the record's - the third assertion about this family's spelling to turn out that way in one day, after the LORENA of Victor Lerena's 1889 margin and the FRUCTUOSO RIVERA printed over the 1831 act's own FRUTUSO RIBERA. The confusion of L, Ll and S in these registers is real and documented elsewhere; these three instances of it were not in the registers at all.

was That JOSEPH PONTECELI and AGUEDA RIVERO, godparents at the baptism of Dominga Josefa Llerena on 15 May 1804, are sourced to "Canelones confirmation records 1804 ... ark 3:1:3Q9M-CS2W-73LL-2" .

now THE ARK WAS THE WRONG ONE, and both rows carried it for a week. That ark is the Canelones CONFIRMATION LIST - a page of one-line entries on film 008165032, opened on 21 September, where neither name appears. Their act is Dominga Josefa's own BAPTISM, on film 008166571 image 01130, and it was read the same day: 'Padrinos JOSEF PONTECELI y [Agueda Rivero], de la misma Villa'. The notes were right and the citation was crossed - two records taken down in one sitting off one search. Corrected to ark 3:1:3Q9M-CS2Z-W9V2-Y, which is the ark Dominga Josefa's own row already carried.

was That the 1833 act naming Sancho FRANCISCO is "a priest's slip" , that "every other record this archive has read off a plate calls him SANCHO" , and that it is "one act against five" - written into the register's Sancho row on 21 September 2026.

now ALL THREE WERE FALSE, AND THIS ARCHIVE'S OWN FILE AND ITS OWN HYPOTHESES PAGE ALREADY SAID SO. The 1846 La Mauricia acts were READ OFF THE IMAGE by this archive and say FRANCISCO; the count is SIX saying Sancho against THREE saying Francisco; and the readings split ACROSS methods rather than along them. The note was written without reading data/sancho-or-francisco.tsv, which had settled the question a week earlier and better - the position being that a DOUBLE FORENAME is the reading that requires nobody to be wrong. WHAT THE DAY ACTUALLY ADDED IS WORTH MORE THAN WHAT IT GOT WRONG: the 1833 act was read off the plate - film 007713376 images 01176-01177, 'Fran.co Lerenas y M.a Fernandez' - so the one Francisco-bearing ORIGINAL is no longer possibly a machine artefact, and the double-forename reading is stronger than it was.

was That the FamilySearch ACCOUNT was signed out and that "David will need to sign in to FamilySearch again" , published on /method/, in data/false-zeros.tsv and on work-list row 184.

now THE ACCOUNT WAS NEVER SIGNED OUT, AND DAVID NEEDED TO DO NOTHING. A sibling archive on this machine ran the record search and the full-text search in the same minutes and got 3,652 and 3,967 with the account name in the navigation bar. Each Claude session's in-app browser is a separate browser with its own cookies: ONE CONTEXT lost its auth and recovered on its own. In a fresh tab the Gonzalez control returned 479,662 and the Avelino-with-Ambrosia query returned 10 again. The symptom was real and the diagnosis of a lost session was right for that browser; the REMEDY was wrong, and it asked a person for something the machine could have settled by opening a tab. ALL THREE HELD-OVER NULLS WERE RE-RUN AGAINST A LIVE CONTROL AND ALL THREE ARE STILL ZERO.

was That 'only ROSARIO in COLONIA is still unaccounted for', written into worklist row 156 on 20 September when that remainder was split out of row 120.

now ROSARIO-COLLA WAS NEVER UNACCOUNTED FOR. IT IS THE ORIGIN. This archive's own notes already read 'RAMON LERENA x MARIA JOSEFA GONZALEZ, of ROSARIO / COLLA, fathering children c.1800-1820' and 'from Rosario-Colla across San Jose, Flores, Florida, Salto'. Colonia is not the department the family failed to reach - it is the one they came FROM, and the 1837 San Jose marriage gives Mauricio's birthplace as COLLA outright. The row was written from row 120's summary line without reading the file underneath it, which is the same fault as the Rosario-parishes correction three days ago.

how Caught while re-measuring Colonia for compound surnames.

was That the 1812 padron 'records four enslaved people in Sancho's household and GIVES NONE OF THEM A NAME'. Carried in this archive's data and on its pages since the padron was first quoted from a published transcription.

now THE PADRON NAMES ALL FOUR: ANTONIO, JOSE, JUANA and TERESA. They are entered after the youngest child Dominga and before RODRIGO FERNANDEZ, the father-in-law, who closes the household - and Antonio is written out as 'negro esclavo'. The claim came from trusting a transcription that dropped them. Four people this family held were nameable all along, and this archive said they were not. Antonio, Jose and Teresa are now in the register; Juana already was, from the baptism act, and is now corroborated by a second document of the same year.

how Read from the image, ark 3:1:3QS7-99ZR-NKNR, 20 September 2026.

was That the four people standing ABOVE Sancho on the padron - Juana 41, Juan 'negro esclavo', Gabriel and Maria - might be his enslaved household, published on /hypotheses/ for about an hour today.

now WRONG FOUR, AND THE WRONG DIRECTION. Those are house 59, a different address. Sancho's four are below his six children. The tell was DOMINGA: the transcription ends the children at her, so everything between Dominga and Rodrigo Fernandez is the rest of the household. The Juana aged 41 in house 59 is a different woman from Sancho's Juana, and saying so matters more than the error, because two Juanas on one page is exactly how a family tree acquires a wrong mother.

how Caught within the hour by reading downward instead of upward.

was That film 007499043 has 'EIGHTEEN ITEM BOUNDARIES', published on /uruguay/ and in the film register within the hour.

now IT HAS EIGHTEEN DARK-FRAME CLUSTERS, WHICH IS NOT THE SAME THING. The brightness scan separates dark frames (16-30) from ordinary pages (140-156) perfectly - but a filming target board is only one kind of dark frame, and so are black book covers and spines. Image 00127, predicted to be a title board, is a cover; image 00091 is a FIN DEL ITEM card with no title on it. Only three of the eighteen have been confirmed by eye. The scan still did its job - it reduced 3,208 images to eighteen places worth looking - but finding candidates was reported as finding boundaries.

how Caught by testing the inference on the first cluster rather than publishing it untested twice.

was That the Dominga of the 1836 Montevideo census was 'a candidate, not a finding' and 'attached to Sancho by nothing' - published on /uruguay/, and a NEW PERSON added to the register for her.

now SHE WAS PROVED ON 14 SEPTEMBER AND I DID NOT LOOK. The register already held DOMINGA JOSEFA LLERENA FERNANDEZ, b. 13 May 1804, daughter of Sancho, wife of CAPITAN DON ANTONIO RUIZ DE CARVALLO, on the Catedral act of 12 May 1827 - which names her as 'Dona Dominga Llerena de Carvallo' and baptises the very LUCINDA who appears beside her in the census. The duplicate has been removed and its evidence folded into the existing person. The sting: one hour earlier I declined to add three children of household 97 precisely because they might duplicate people already here - and then created exactly that duplicate by not checking. The census and the 1862 Rio de Janeiro marriage remain genuine additions.

how Caught while chasing which parish register rows 167 and 168 were blocked on.

was That the Gilberto Lerena who landed at Buenos Aires on 27 February 1928 was 'Not in the 1895 census household of Gilberto (1854)' - stated as fact in the register, and the only thing holding his row apart from another row in the same register.

now HE MAY WELL BE IN IT. Line 6 of that schedule reads "Salvanach Lerena, GILBERTO", varon, 17, soltero, R. Oriental - the eldest son, born 1877 or 1878 against the arrival's 1879, already in this register as Gilberto LERENA Salvanach (the younger). The schedule had been read as a household on 14 September and the objection was never checked against it. The two rows are now cross-referenced and the test named; they are NOT merged, because both are index-level.

how The duplicate gate flagged the pair, and working the backlog meant reading the household file that the objection rested on

was That the baptism of ALEJANDRO at La Mauricia dates him to 29 May 1843, and that this 'pushes their marriage back before May 1843' - the marriage of Avelino Lerena and Ambrosia Traibel, stated in the register since 14 September.

now THE YEAR ON THAT PAGE IS A FORMULA. Read end to end at magnification on 20 September, all three acts of 7 January 1846 give the birth year in the identical words, "del pasado ano cuarenta y tres" - and the third is for a different couple. The phrase contradicts itself, since the acts open "En siete de Enero de mil ochocientos cuarenta y seis" and "el pasado ano" is then 1845. And it cannot be literally true: the two children of Luis and Justiniana on the page would be born five months and ten days apart. Nothing outside the page dates Alejandro, so his birth is undetermined between 1843 and 1845 and the safe floor for the marriage is before May 1845.

how Reading the page for a different reason - chasing why a father and son both carried 9 February as a birth day - and magnifying the numerals instead of trusting two transcripts already in the archive

was That the 1843 reading was retracted on 14 September and the archive moved on. Five data files went on asserting "9 February 1843" for six days, one of them arguing for it in a paragraph headed WHAT THIS SETTLES, and the correction's `retired` column was left empty so no gate enforced it.

now A CORRECTION THAT IS NOT CARRIED BACK TO THE FILE THE READING CAME FROM IS NOT A CORRECTION. All five now say 1845 or quote the old reading; correction 50 has its `retired` column filled, and check-corrections.py enforces it. Also withdrawn: "Eleven months before Luis Eugenio was born on 9 February 1843. The fit is exact." - the marriage of 3 March 1842 is followed by Justiniana Mauricia, sixteen months later, and Luis Eugenio is the second child at thirty-five months.

how Grepping data/ for the retracted string while chasing an unrelated date

was That work-list row 111 read "TAB MHG 1095/48 - tell Anne before Tuesday, its death notice is already in the house", with every limit on it kept in the note. And that the note then said "the other four on her list" while listing five.

now THE ROW WAS ACTIONED AS "STOP", AND THE WHOLE BOOYZEN PULL AT PRETORIA WAS CALLED OFF. The finding was that ONE document of the six-estate list is already in the house - the death notice of 1095/48. Cancelled with it were the will and liquidation account of that same estate, which nobody holds, and five entire estates - 3734/46, 769/52, 5155/51, 10/57 and 52653 - none of which is in the folder. A row whose headline reads as an instruction and whose limits live in a note is a badly written row, and this archive wrote it. The headline now states the scope, the miscount is fixed, and the row says what it cost.

how David asked why Anne had been told to stop, which did not match what this row actually found

was That ALFREDO JUSTINIANO LERENA TRAIBEL was born in March 1858, and that "his baptismal age of 1858 sits oddly against the 35 he gave at his 1892 marriage". The register blamed the discrepancy on the record and on "ages at marriage", the weakest date in genealogy.

now BORN 2 OCTOBER 1856, ON THE ACT THIS ARCHIVE HAD ALREADY READ. Re-read off the image on 20 September - film 007713686 image 2847 - it says "que nacio el dia DOS de OCTUBRE ... cincuenta y seis". The baptism date had been written into the birth column. AND THE AGE WAS NEVER ODD: a man born 2 October 1856 is exactly THIRTY-FIVE on 3 February 1892, which is what he told the priest. The record agreed with itself the whole time.

how A FamilySearch search run for an unrelated reason printed "Birth 2 October 1856" beside the christening of 12 March 1858

was That ANA ROSARIO LERENA TRAIBEL was born in March 1858 and was 'probably twins' with her brother Alfredo Justiniano, both baptised at San Francisco de Asis on 12 March 1858.

now BORN 14 JANUARY 1855 - and they are twenty-one months apart. Her own act, on the same opening, reads "que nacio el dia CATORCE de ENERO ... cincuenta y cinco". She was carried to the font at THREE YEARS OLD and her brother at seventeen months. "Probably twins" was an inference from a shared baptism date and the acts refute it - this family baptised in batches, years late, and did it at La Mauricia in 1846 as well. The same inference is attached to the two MORATORIO girls of 1880 and is now flagged there as untested.

how Reading the second act on an image already open for the first

was That this register held THREE separate people for Carlos Augusto Federico Lerena Salvanach, TWO for Avelino Carlos Lerena and TWO for Rosa Maria Eppens - and that data/relations.tsv gave Carlos a wife and a son the register had already ruled out.

now SEVEN ROWS WERE FOUR PEOPLE. The archive had done every identification and none of the merges: one note read "two independent sources, same couple", another "see the CEMLA row for the same man", a third "the 'Rosa M. E. Lerena' of the CEMLA list". The arithmetic confirms the hardest of them - the Retiro act of 25 March 1914 calls him "de treinta anos", and a man born 9 July 1883 is exactly thirty that March. AND THE EDGES WERE WORSE THAN THE ROWS: relations.tsv made him the husband of ADELA CAMIGLIA and father of AVELINO CARLOS at Mendoza, which the Rosa Maria Eppens row had explicitly denied since 14 September - "two men, one name, and the register now says so on both rows." It said so on the rows and nothing told the edges. The Mendoza man now has a row of his own, and they point at him.

how Working the duplicate backlog pair by pair, which is what that backlog is for

was That lines 10-13 of Gilberto Lerena's 1895 census schedule give the ages '6, 6, 5, 2' - published in data/1895-census-gilberto-household.tsv and used to keep three register rows apart.

now THEY ARE 8, 6, 5, 2. MARIA A. IS EIGHT. Read off the image at magnification on 20 September, film 004149111 image 8. AND THAT ONE DIGIT SETTLED TWO IDENTIFICATIONS. At six, Maria A. could not be MARIA ANGELICA JULIA LERENA SALVANACH, baptised 31 October 1886; at eight she is exactly that. MARIA C., six, is MARIA CARLOTA, baptised at about a year old on 29 October 1890 - and her nacion column reads ARGENTINA, which is what that baptism row has always claimed for her. Both are merged, with LUIS F. and CARLOS A. earlier the same day. The enumerator wrote initials and the priest wrote names, and this archive held four of these children twice.

how Working the duplicate backlog: two pairs were blocked on a single digit, so the image was opened

was That 23 Geneva Place, Bideford and North Street, Northam were "NOT TESTABLE FREE". Written into data/devon-addresses.tsv and worklist row 123 this morning, on the strength of FreeCEN holding neither 1871 nor 1881 for those parishes.

now THE CENSUS SENTENCE WAS TRUE; THE VERDICT ON TOP OF IT WAS NOT. One route was tested and the verdict was written as though all had been. The UK National Archives had never been searched at all, and Geneva Place turns out to be a council school whose log books and ADMISSION REGISTERS survive for 1885-1949 (North Devon Record Office 629C; also Kew ED 21/26168 and ED 70/427). Admission registers give the child, the date of birth, the father and the address. A null must name the routes it tested, and this one silently claimed a route it had not tried.

how David asked whether this archive had ever looked at nationalarchives.gov.uk. It had not.

was That 'the other Rosario parishes' were a live route to Pablo Armando's 1882 baptism. Offered as one of three remaining routes, and as the second-best of them.

now THERE WERE NO OTHER PARISHES IN 1882. Of the five Rosario parishes, four open in 1888, 1896, 1899 and 1899 - all after his birth. The Catedral was the only parish in existence, and its register is the one already walked act by act with no Lerena in it. The archive's own rosario-five-parishes.tsv said so in a column headed 'does it reach February 1882?', and I recommended the route without reading it.

how Checked when the route was actually walked, an hour after recommending it.

was That a birth in February 1882 'should have been registered civilly whether or not he was baptised'. Written into worklist row 152 and given to David as the main line on the central question - the one thing left worth doing.

now CIVIL REGISTRATION IN SANTA FE BEGINS IN 1886. Four years after he was born. The act does not exist and no letter will produce it. The route is not dead, but the ask had to be rebuilt: a late inscription (inscripción tardía), and above all THE PARENTS' DEATH ACTS, which would have been registered because Juan Carlos and María would have died after 1886 - and an Argentine acta de defunción names the parents and spouse of the deceased. The corrected route is stronger than the one it replaces.

how Checked before drafting the letter, because a wrong premise would have wasted the request.

was The blood-relatives chart, as published, drew a family larger than the family is - two Juan Carlos Lerenas, two Maria Lerenas, two Mary Septima Taylors, and Pablo Armando standing beside himself labelled 'aunt or uncle'

now Nobody was duplicated in the data. The drawing was. A husband or wife was always drawn as a chip beside the person, with no check that they were already on the chart in their own right - and every married pair of ancestors is two blood relatives, so each got a box AND a chip of the other. Nuno Fernando's 13 relatives came out in 22 boxes Separately, this register deliberately keeps a row for the mangled machine-index reading of Pablo Armando - 'ARMAND[O] LORENA (as machine-indexed) = Pablo Armando LERENA' - so that the false trail stays visible, and the sibling pass handed that row its own person as a brother. A chip is now drawn only for somebody the chart does not already hold, and an alias folds onto the person it is a spelling of The same chart is now 14 boxes for 12 people. This was never a Lerena-only fault: counted across every chart in every archive in the estate, 218,856 duplicate boxes were being drawn, worst in the largest families - Mazza's chart for Frank Mazza drew 232 of its 530 people twice.

how David looked at the page and said it made no sense, naming the repeated people. He was right, and it had been published in that state since the page was built. No checker could have caught this: every gate in this archive reads data, and the data was correct It was the picture that lied.

was That a granddaughter of Pablo Armando carries "no Spanish component whatever", and that her Italian is "northern-dominant" - the signature of the Ligurian and Piedmontese emigration to Rosario

now Both sentences were written from ONE company's reading. On 16 September 2026 the same saliva was read at 23andMe - the account the MyHeritage kit was uploaded from in 2018 - and it disagrees twice. It returns 2.1% Spanish and Portuguese where MyHeritage returned 0.0%, so the flat zero was a fact about a reference panel and not about her DNA. And it puts her Italian in SICILY, where MyHeritage put it in the north. The northern reading is withdrawn as a DNA argument; the Ligurian names in the Rosario register are still in the Rosario register, but they were never a DNA claim. What survives, and is stronger than before: both companies find Italian - 11.1% and 8.0% - in a woman whose other three grandparents are English and Cape Afrikaner. One company's zero has been replaced by two companies' agreement, which is a better foundation than the one the hypothesis was built on.

how Caught by doing the obvious thing that had never been done: reading the same DNA at the second company that already held it. Nothing new was tested and nothing was bought. The second reading had been sitting in an account this family already owned since 2018.

was Nothing was published from this, because the control caught it first

now A battery of twelve surname and place probes was run against the 1,500 DNA relatives at 23andMe - Lerena, Llerena, Lerina, Argentina, Uruguay, Montevideo, Rosario among them - and every single one returned ZERO. Before any of them was written down, two controls were run in the same box: BOTHA returned 0, and SOUTH AFRICA returned 0. The facets on that same screen show thirty Bothas and 581 relatives with a South African grandparent. The box searches display names and the user's own notes, and nothing else. Twelve nulls were one step away from being published as findings, and the control rule stopped all twelve. Recorded here as a correction to nothing, because the rule working is worth more on the record than the rule existing.

how The control-before-null rule, doing exactly the job it was written for.

was A ranked list of archive files to buy at Pretoria, whose top two entries were already in the house

now This archive drew up a pull list for a researcher going to the National Archives on 22 September, at R2,850 for the first file, and ranked two Transvaal death notices first and second. Both had been photographed and saved in David's own folder - Genealogy & Family History/Lerena - since November 2021. Nobody had ever opened it. Worse than the duplication: the rationale for the second was that it would NAME NUNO FERNANDO'S PARENTS, and item 19 of that notice carries the single word "Deceased" and a ditto mark. No names at all. The first was ranked to establish Rogue Luis Armando's place of birth; the notice says CAPE TOWN, and had been saying so for five years. Both are struck from the list, which now leads with two divorce files. The same folder also held the 1911 marriage register, which was ON THE WORK LIST as a thing to go and find.

how Caught because David asked, in as many words, whether this archive had searched the folder. It had not. The archive had a register of what it had READ and no register at all of what it already HELD. A rule is now on /method/: before any list of documents to acquire, list the documents already held.

was That the death notice of Pablo Armando, KAB MOOC 6/9/17017, was the one document that could name Juan Carlos Lerena - and that it had to be fetched from the Western Cape Archives

now Both halves were wrong, and in opposite directions. It did not have to be fetched: a photograph of it had been in David's own files the whole time, while an errand requesting it sat open at the Western Cape Archives. And it does not name them. Item 3 reads "Father: not known. Mother: not known." The informant was ROSE LILIAN THEYS, who signed as "Present at death" and "Executrix Nominate" - the woman he lived with at No. 7 Forth Road, described in the liquidation account as a spinster and NO RELATION. She could not name his parents, and none of his three living adult sons supplied them either. The archive had built a fortnight of priority around a document whose answer was a null and whose location was a folder. One consequence stands: the 1908 Argentine enrolment card is now, definitively, the ONLY document on earth naming Juan Carlos and Maria Lerena.

how Found on the same sweep that David prompted, when he asked whether the whole of his Documents folder was worth scouting. It was. The archive had never run a filename search for its own subject's name across the disk it was sitting on.

was That the family had no memory of Pablo Armando's origins beyond "Argentine"

now They had a surname, and it is not Lerena. Sue Rowbotham's Taylor family history, written in England and shared with this family in 2020, records that Mary Septima "went out to South Africa to work for a widower by the name of ROBERT (or ROBERTO) MARINO". The three sons she names - Louis, Cardo and Nuno - are exactly right, so the memory is anchored in real people; it is the surname that has drifted, or that was never Lerena. MARINO is one of the commonest surnames in Italy, and this archive has argued from two companies' DNA readings that he carried Italian ancestry. The two lines of evidence are wholly independent and neither knew of the other. This is corroboration, not proof, and it is recorded as the former.

how Read from a Word document in the Taylor folder, on the same sweep. The archive had held it, unopened, for as long as it had held everything else.

was That the 1950 records agree with the passport about how old he was

now They do not, and the disagreement is nine years. The Argentine passport and the 1908 enrolment card both give 22 February 1882, which makes him 68 when he died on 19 March 1950. The death NOTICE says "77 years - approx". The death CERTIFICATE says "76 years". Both South African documents put him in his mid-seventies, and they agree with each other. Neither was completed by anyone who knew him: the notice by Rose Lilian Theys, no relation, and the certificate by W. Fraser, an undertaker whose stated qualification is "causing burial". The simplest reading is that strangers guessed high, which is what strangers do - and it is recorded rather than smoothed, because in the same folder somebody had filed a Buenos Aires baptism of 1872 for a ROBERTO JUAN ROQUE LLORENS, whose age would fit.

how Found on the folder sweep. The archive had transcribed the death notice's other fields this morning and had not stopped at the age.

was That the 1872 Buenos Aires baptism of Roberto Juan Roque Llorens was pulled deliberately by the family, as a rival identification they were pursuing

now Wrong, and it was an inference this archive had no business making. The image sat in a folder with no note attached, and this archive wrote that "somebody pulled that deliberately". David says it arrived unsolicited, in the mail, from a researcher in Argentina, and that he regards it as likely no relation. The document is kept and the working is kept, because the archive had already reached the same verdict from the documents before he said so - and because the name has leaked into the MyHeritage family tree, where he is entered as "Roberto Paul (Llorens) Lerena", so the reading is in circulation and is better refuted in public than left to drift.

how David said so, plainly, when he read the page. The lesson is narrow and worth keeping: a file with no note attached tells you nothing about why it is there, and this archive invented a motive to fill the gap.

was That the Bailey cluster on the DNA chart is "probably" the BRAYLEY line - Mary Septima Taylor's mother's family

now Written this morning on a surname resemblance, and tested this afternoon. It does not stand. The real Brayley matches were found: "Lisa" of Australia at 23.0 cM sharing BRAYLEY and BAKER, and "Ron Brayley" of the United Kingdom at 16.2 cM sharing BRAYLEY. Neither is in cluster 14, and none of the three BAILEY matches in that cluster carries Brayley as an ancestral surname at all. The Brayley signal sits at 14-25 cM in Britain and Australia; the Bailey cluster sits at 61-116 cM in the United States. Different signals, different distances, nothing joining them. BUT THE ABSENCE IS NOT EVIDENCE EITHER: the Bailey trees hold TWO and FOUR people, and MyHeritage cannot compute a shared surname from a tree with two people in it. So the correct statement is that the link is UNTESTED, NOT REFUTED - and "probably" was more than the evidence carried. The one thing that would settle it is a message to the 116 cM match asking for two generations of names, and that is David's to send.

how Caught by running the test instead of admiring the coincidence. The GEDCOM handed over the actual Brayley surnames - Brayley, Reed, Wybourn, Temple - and the moment they were put through the match search the Bailey cluster failed to appear among them.

was Everything this archive published from the GEDCOM it found on the disk - the Brayley ascent, Anton Armando's parentage, Careen Marie's dates, and an Australian birth at Dapto, New South Wales

now The file was `2018-07 Defranceski.ged`, an eight-year-old backup. David: "Do not use the 2018 gedcom. That is an old backup only. And is incredibly wrong and out of date." He was right, and the current export - 15 September 2026, 15,643 individuals against the backup's 8,286 - contradicts it point by point. THE AUSTRALIAN BIRTH DOES NOT EXIST: the backup put Paul Lerena at DAPTO, WOLLONGONG, NSW on 3 November 1951; the current tree has Paul Raymond Lerena, born 11 March 1951 at Johannesburg. This archive had flagged that entry as "a date that cannot be true" because his sister was born in Johannesburg six months later - the flag was right and the cause was the backup. CAREEN MARIE was published as born and died 1945; she was born June 1947 and died 22 June 1947, which is what Cheryl told her English cousin in 2020. The family were right and the backup was wrong. And the backup carried him as "Roberto Paul (Llorens) Lerena, born EST 4 SEP 1872, Argentina" - the unsolicited Buenos Aires candidate baked into his name, his birth date AND his parents. The current tree has restored Pablo Armando / Robert Paul Lerena, born 22 February 1882 at Rosario. All of it is withdrawn and re-derived from the current export.

how David said so plainly. The lesson is one this archive should not have needed: a file's DATE IS PART OF ITS PROVENANCE, and "found on the disk" is not a source. It was treated as a source because it was large and internally consistent, and it was neither current nor right. The two errors this archive DID catch by internal arithmetic - the impossible Dapto date and a man dead at 117 - should have been read as symptoms rather than as isolated slips.

was Nothing yet, and this is a warning rather than a retraction

now The CURRENT tree still carries Pablo Armando's parents as "Roberto Llorens" and "Francisca Fernandez Lerena" - the Buenos Aires couple from the unsolicited Argentine candidate - while his birth has been corrected to Rosario, 22 February 1882. Those two cannot both be true: the 1872 Buenos Aires baptism is the only thing that ever attached those parents to him, and it is the same document David says is likely no relation. The birth was fixed and the parentage was left behind. Recorded here so that it is fixed in the tree before it is exported again, and so that no future session mistakes it for evidence. The mother's surname "Fernandez Lerena" appears in no register: the Buenos Aires act says FRANCISCA FERNANDEZ.

how Noticed while re-deriving from the current export, by reading the parents field after the birth field instead of stopping at the good news.

was That the enrolment card's SERIE field reads "3885". Published on /enrolment-card/ in the table of what the card says.

now It reads "E 3555". The letter prefix was dropped entirely and two digits were misread. The SECCION field, "I 3322", was right - and it has the same shape, a letter and four digits, which is what should have prompted the doubt: a field transcribed WITHOUT its letter, sitting directly above one transcribed WITH it, is an internal inconsistency this archive should have seen before David did.

how DAVID READ THE ORIGINAL. He was asked what the card was telling him, looked at it properly, and sent back both fields. Neither number carries any weight in any argument here - no conclusion moves - but it is logged because a transcription is the one thing this archive asks readers to trust, and a wrong one is worse than a missing one.

was That on /direct-line/, "Above him there is nothing. His parents are unnamed in every document held here." Published, and left standing for six days.

now His parents ARE named. The enrolment card gives Juan Carlos Lerena and Maria Lerena, and has done since 10 September 2026. The same page said so in its own opening paragraph - "an Argentine army enrolment card... names Pablo Armando's father and mother. For months there was nothing above him. There is now one generation" - so the page contradicted itself from top to bottom. Now reads "Above him there is ONE GENERATION AND NO MORE", which is the true and still-bleak position: two names, given by their son, and no other record of either anywhere.

how Caught while rebuilding the front page at David's request. Nobody went looking for it. This is what a big find does to an archive: the good news gets written at the top of every page and the old despair stays at the bottom, and the two sit there disagreeing until someone reads the whole page in one go. Graded SERIOUS, not minor: it is not a slipped digit, it is the archive telling a reader the opposite of what it knows, on the page that exists to state the line.

was That work-list row 74 said "the 1908 enrolment card puts him still in Argentina." Carried on the work list, unchallenged, while the search for his passage out was framed around it.

now THE CARD SHOWS THE OPPOSITE. Its *Oficina enroladora* reads CAPETOWN and both *Region Militar* and *Distrito Militar* are BLANK - the signature of a man registering at his own consulate because he was ALREADY ABROAD. The card is evidence he had LEFT Argentina, not that he was in it. And the 1908 is unsourced besides: the card bears no year but the class of 1882, sits in a collection catalogued 1911-1936, and the passport gives a Cape Town consular libreta of 1915.

how Caught while working row 136. The error mattered more than a wrong fact usually does, because it set a FLOOR on the search: if he was in Argentina in 1908, his departure is after 1908 - and rows 74 and 108 were hunting a 1901-1905 window that the same archive had derived from his son being born at the Cape in 1905. The work list was holding two incompatible positions and neither page noticed. Graded serious: a wrong fact on a work list is worse than a wrong fact on a page, because a work list tells the next session where to look.

was That Pablo Armando's absence from the 1895 National Census, at thirteen, suggested he had already left Argentina. Written on this archive's own work list earlier the same day.

now It suggests nothing of the kind, and the archive should have known better within the hour. One person can always be missed or mangled by an index - and THIS person's name demonstrably mangles. The Argentine army's own index turned his enrolment card into "ARMAND LORENA", born "Rosario de Lerma, SALTA". A boy who can become Armand Lorena on one index can be anything at all on another. The absence is consistent with having left and equally consistent with sitting in Rosario under a spelling nobody would think to search.

how Self-caught, about forty minutes after writing it, while running the control on the census. The control is what separated the two claims: a GONZALEZ search returns 88,739 records with the 1895 census reaching Rosario itself, so the SURNAME being absent from Santa Fe province is strong and the INDIVIDUAL being absent is weak. Same census, same search, two completely different evidential weights, and it took a control to see it.

was That FLORES, SAN JOSE, FLORIDA and ROSARIO-COLONIA were "four Uruguayan departments this archive has NEVER SEARCHED", and that the Bernabe Lerena / Casimira Pelaez family of Trinidad was a find. Written into data/gedcom-import.tsv, then into work-list row 120, then into data/uruguay-four-departments.tsv, and stated to David in those terms.

now THE ARCHIVE HAD SEARCHED FLORES THREE DAYS EARLIER AND CALLED IT THE FOUNDING NEST OF THE SURNAME. `trinidad-flores-kindred.tsv`, read 13 September 2026, is titled "TRINIDAD, DEPARTMENT OF FLORES, URUGUAY - the founding nest of the surname" and holds eleven children, the Basque mothers BELOQUI, ECHEVERRY and PELAEZ, and a correction of its own. `lerena-pelaez-trinidad.tsv` maps four Lerena-Pelaez consanguinity dispensations there. BERNABE LERENA AND CASIMIRA PELAEZ ARE IN IT BY NAME. San Jose, Florida and Colonia have no dedicated study, so "never searched" was merely too strong for them; for Flores it was simply false.

how SELF-CAUGHT, about an hour after publishing it, and only because the Flores work led back into `the-three-alejandros.tsv` which cites "the six Trinidad, Flores fathers". THE MECHANISM IS WORTH MORE THAN THE ERROR: the false claim entered from the GEDCOM import, where a note about what the TREE offered was written as a claim about what the ARCHIVE had done - without checking the archive. A statement of the form "this has never been searched" is a claim about 161 data files, and it must be grepped before it is written, every time. WHAT SURVIVES: the record counts, which are new and real - Flores 608, San Jose 315, Florida 231, Colonia 189, against Canelones 191. A nest has been read; a department has not.

was That DGS 7713386 is 'Indices de bautismos, matrimonios y defunciones, 1726-1800', and that opening it under L was 'THE CHEAPEST TEST IN THE ARCHIVE'

now Two images were sampled on 15 September 2026. Image 300 is a burial register of the Catedral Basilica de Montevideo for 1887; image 1500 is a numbered burial register for 1905. Neither is an index and neither is eighteenth century. Two samples out of 2,244 images do not prove the index is absent from the film, but the attribution is not confirmed and the errand must stop calling it cheap or certain.

how The attribution came from a catalogue reading, not from the film. The film was never opened until now.

was That the castrense chaplain was 'Juan Miguel de Mendoza', and that a neighbouring 1789 entry named an enslaved man belonging to a Llerena

now Both were misread at fit-width and corrected at full magnification before either reached the site: the chaplain is JUAN MIGUEL DE BERROETA, and the slave-owner is FRANCISCO JUAREZ, not Llerena. Recorded because the second misreading would have manufactured a Llerena in the regiment's own book - the exact thing this archive wants to find, which is precisely why it nearly got through.

how Caught by re-reading at magnification, as the standing rule requires.

was That the second column of data/sa-deceased-estates.tsv was a death_year

now It is the year the estate was ADVERTISED in the Government Gazette, and the file called it death_year for all eleven rows. Where the J193 has now been read the two differ by years: Ricardo Juan Carlos died 21 December 1994 and was filed under 1996; Anton Armando died 3 April 2003 and was filed under 2013. Column renamed to gazette_year and the true dates recorded.

how Caught by reading the actual gazette notices instead of the index that summarised them.

was That gendatabase and this archive contradicted each other about Anton Armando Lerena, and might describe two men

now There was no contradiction and there is one man. This archive's row read '2013 ... 68', and this session read that as died 2013 aged 68, implying birth about 1945 - but 2013 was the gazette year and 68 was the PAGE NUMBER, both of them columns in the archive's own file. The J193 says: Anton Armando Lerena, born 1952, died 3 April 2003, estate 15127/2013 advertised on page 68 of the gazette of 16 August 2013. gendatabase was right.

how Caught by fetching the gazette page that both records were pointing at all along.

was That DGS 8166573 held nothing below image 1057

now Written this morning from three probes - images 1, 500 and 1000 - all of which happen to fall in the same gap. The film has TWO blocks: images 72-478 and 1057-3295. Block one is the Canelones baptism register 1776-1794, a different item entirely, and this archive already held an ark into it. The claim was not wrong about the images it tested; it was wrong to generalise from three of them.

how Caught by resolving an ark the archive already had, which landed at image 274 - inside the range this session had declared empty.

was That three images found by the parent-pair query were unread by this archive

now All three were already read and recorded, and in fuller words than the index gives - Dominga Josefa's baptism of 1804 with its godparents, Francisco Juan Ramon of 1826, and Lucinda Gregoria of 1827 with the Capitan, the Mayor and a General as godfather. The claim came from inferring an unread ACT from an unread COLLECTION, which does not follow: the same act reaches this archive under different arks by different routes.

how Caught by grepping the archive for the names before writing them up as new.

was That the 1791 act created 'two Lerena-Aguilar unions in Cordoba' worth reopening the 1754 Granada file for

now Only the 1754 groom is Cordoban. His bride, Ana Maria de Aguilar, is of La Mala - La Malaha, in GRANADA province - which this archive had already recorded. Our Rosalia Aguilar is of Cordoba. So the convergence is one very common Andalusian surname in two different provinces, not a family. An earlier session had already run the decisive test - a structured search for a son Sancho de Lerena y Aguilar, returning a controlled No Results Found - and was right to close the file.

how Caught by opening the record and finding the archive already held every field on it, including the bride's birthplace.

was That Sancho Llerena's last certain appearance as a living man was the deposition of 9 January 1798

now Written twice this session, and wrong. The 1812 PADRON is later and is an enumeration of the living: it returns him at calle San Juan No 60, household 59, as Sancho 'andaluz', 45. So the last document that shows him alive is 1812, not 1798, and the unaccounted gap before the trees' death date of 1832 is twenty years rather than thirty-four.

how Caught by listing all fourteen records that name him and marking which ones require him to have been standing there.

was That the Rosario confirmations of 1890 were taken sex by sex, so a boy would be in a separate run of pages

now Wrong, and built out of eleven rows. The first eleven entries on the first page read happened to be girls, and I turned that into a structure and a search strategy. Entries 133, 135 and 138 on a later page are Domingo, Santiago and Pablo - all "hijo leg. de" - mixed straight in among the girls. There is no boys section to find and no shortcut: the whole list has to be read. The ages run from two to fourteen in one sitting, so age rules nobody in or out either.

how Reading a second page of the same register a few hours later, instead of leaving the first page as the description of it.

was That the first wife's surname read as something like "Dille" or "Dulce"

now CAROLINA OLIVA. The 1842 act names Luis Lerena "viudo de Dona Carolina" and I could not read the surname confidently off that page, so I wrote down what it looked like and said it was not confidently legible. The marriage petition of 3 September 1836, found an hour later, gives it plainly: "Dona CAROLINA OLIVA, tambien natural de esta Ciudad". The uncertain reading is left recorded rather than deleted, because the next person squinting at that word should know what it is not.

how Searching for the marriage instead of squinting harder at the word.

was That the Sancho question was closed - "three originals against two copies is not a conflict"

now REOPENED WITHIN THE HOUR BY AN ORIGINAL ACT THAT SAYS FRANCISCO. I argued that Luis Lerena's father is Sancho on three original acts, against two 1846 entries reading Francisco which say of themselves that they are later transcripts - and concluded the conflict was only copies against originals. Then the Catedral baptism of Justa Pastora Lerena's second child, October 1833, turned up naming her parents "FRANCISCO LERINA y Maria Fernandez". That is an ORIGINAL. The tidy argument does not survive it. What IS constant across every record and both spellings is MARIA FERNANDEZ, the wife. A double forename - Sancho Francisco - would explain everything without anyone being wrong, and is now the leading reading. Sancho's own marriage act, never opened, would settle it.

how Following a new sister, Justa Pastora Lerena, to her second child - and reading the grandparents line instead of stopping at the one that agreed with me.

was That Don Luis Lerena's first wife was CAROLINA OLIVA

now CAROLINA ORIBE. Recorded as Oliva twice on 14 September - once to correct an earlier "not confidently legible" reading, and stated as settled. The source of the error was FamilySearch's index of the 1836 petition, which says Oliva; I read the image expecting it and saw it. A Catedral baptism of 4 July 1837, in another book and another hand, names her "Dna CAROLINA ORIBE" as godmother beside her husband. The petition, re-opened and magnified, reads ORIBE too. AND THIS ARCHIVE'S OWN HYPOTHESES PAGE HAD CARRIED, FROM A SECONDARY TREE, THAT HE MARRIED FIRSTLY A DAUGHTER OF AN ORIBE - it was right, and I never checked my reading against it.

how Finding a second act that names her, while looking for children of that marriage.

was That the Montevideo dispensation register holds EIGHT Lerena dispensations, every Lerena in it marrying out

now A NINTH WAS FOUND, and the search that produced the eight was structurally blind. Jose Pedro Massera petitioned on 19 June 1913 to be dispensed one conciliar proclama in order to marry "Dona EMA SIXTA LERENA JUANICO, feligresa de la metropolitana" - our own Ema Sixta, eleven days before her wedding. The search formula was "LERENA FELIGRESA", which assumes the parish follows the surname immediately. Her COMPOUND surname sits between the two words, so the phrase could never match - and every child of both Montevideo couples carries a compound surname. The conclusion that no Lerena-Lerena dispensation exists is not withdrawn, but it now rests on a search known to miss exactly the people this archive cares most about, and needs re-running.

how Searching the COMPOUND surname "Lerena Juanico" in full text - the same move that had just turned up her brother Julio as a Juez de Paz.

was That the 1895 census shows a Lerena woman married to a Lerena man - "the pattern the whole search turns on"

now OVERSTATED, AND WITHDRAWN AS EVIDENCE WITHIN THE HOUR. The schedule does read "Lerena de L. Josefa", and the "de L." is formed exactly as on the two lines above - that part was read correctly. The INFERENCE was too confident: "de L." can equally be the enumerator shorthand for THIS HOUSEHOLD, written against every married woman in the Lerena house. Uruguayan civil registration then turned up a JOSEFA LERENA married to FERNANDO MORATORIO, whom FamilySearch's tree attaches to Josefa Lerena Traibel 1851-1947 - the same candidate. If that is her, she was "de Moratorio" and the census tells us nothing about a Lerena-Lerena marriage. The claim was published as "the sharpest question in the project" on the strength of an abbreviation.

how Searching for the marriage I had just said would settle it, instead of stopping at the reading that suited me.

was That the Montevideo dispensation register holds eight Lerena dispensations (revised earlier today to nine)

now A TENTH, found hours after the ninth, and missed for a different reason again. Alfredo Lerena and Hermenegilda Carolina Gavarro, Catedral Basilica de Montevideo, 3 February 1892, dispensed of consanguinity in the third degree with atingencia to the second by Archbishop Mariano Soler personally. The sweep missed it TWO WAYS AT ONCE: the groom is written GERENA - a seventh spelling for this family, in a sentence where his own father is written LORENA - and the dispensation clause comes BEFORE the names, so the formula "Lerena el impedimento" could not match. The count of eight is certainly low, and the file now says so. What survives is narrower: no LERENA-LERENA dispensation has been found, and this family did marry kin.

how Searching the rarest name in the branch - AMBROSIA TRAIBEL - instead of the surname.

was That Avelino Lerena is not shown to be Sancho's son, so the Lerena Traibel branch does not attach

now He is, and it does - settled the same day on a parish act. Two baptisms of 12 March 1858 at San Francisco de Asis name the paternal grandparents of Avelino's children as SANCHO LERENA and MARIA FERNANDEZ, twice on one page. So the whole Lerena Traibel line descends from Sancho - including several people already floating unattached in this register: Mercedes Micaela and Clorinda in the dispensation table, and JUAN ARTURO, the man this archive conflated with another Arturo in its first week and had to retract.

how Searching AVELINO with AMBROSIA - the couple, not the surname.

was That Alfredo Justiniano Lerena's middle name echoed JUSTINIANA LENGUAS, connecting the two branches

now I noticed the echo and flagged it as worth testing. It is nothing: he is named for his GODFATHER, JUSTINIANO LAVALLEJA, who is on the same line of the same act. A tidy connection between the branches that would have been entirely wrong, and the act that raised it also killed it.

how Reading the whole act instead of stopping at the name.

was That ALEJANDRO, baptised at La Mauricia on 7 January 1846, was "A SIXTH CHILD of Luis Lerena and Justiniana Lenguas", born 29 May 1845

now HE IS A SON OF AVELINO LERENA AND AMBROSIA TRAIBEL, born 29 May 1843. The act reads "hijo legitimo de Avelino Serena y Ambrosia Traibel. Paternos Francisco y Maria Fernandez. Maternos Juan e Isabel Centeno. Padrinos LUIS [L]ERENA y JUSTINIANA LENGUAS." Luis and Justiniana are his GODPARENTS - which is how this archive read them as his parents. Both the branch and the year were wrong. He is the ELDEST KNOWN CHILD of Avelino and Ambrosia, and pushes their marriage back before May 1843.

how Reading the full-text transcript of the whole page rather than the entry. The image had been opened on 13 September and the act read too quickly - the parents and the godparents sit in the same sentence, and the archive took the wrong pair.

was That the register contained no contradictions - "245 people checked, nothing contradicts itself", printed on every build for a week

now THE CHECKER COULD ONLY SEE 31 OF 246 PEOPLE. The sibling-gap test reads a STRUCTURED parent field that only 12.6% of the register carries; for everyone else it silently found no siblings and passed. The contradiction it was written to catch was sitting in the data the whole time - Luis Eugenio b. 9 Feb 1843 and Justiniana Mauricia b. 30 July 1843, full siblings five and a half months apart. The file's own docstring names that pair as its reason for existing.

how Reading the 1846 page for another purpose, noticing the two dates, and then asking why the checker had not said so. A guard that reports a clean bill of health over 12.6% of the data is worse than no guard: it manufactures confidence. The checker now prints its coverage and falls back to compound surnames.

was That Luis Eugenio Lerena Lenguas was born 9 February 1843, on the authority of his own baptism act - published since 13 September

now 9 FEBRUARY 1845. The act is a late copy from a legajo, and it is contradicted twice: his sister Justiniana Mauricia's act, the entry immediately above his in the same hand, gives her birth as 30 July 1843 - five and a half months before him; and his civil death registration of 24 December 1901 gives his age as FIFTY-SIX, putting his birth at about 1845. The clerk wrote "ano cuarenta y tres" in the entry above and again in the entry below. A year carried down a column is the ordinary way this happens, and 1845 is the only reading under which all three records stand.

how The consistency checker, once it was fixed to see people without a structured parent field. It flagged the five-month gap on the first run after the fix. The archive then went looking for a third record to break the tie rather than choosing between the two.

was Three unpaired emphasis markers written into the register in one session, each caught only after a build had rendered the literal characters onto a published page

now The slip is always the same shape: closing one emphasis and opening the next with a SINGLE marker where two are needed. consistency.py now checks the TSVs themselves for an odd marker count, before any page exists. On its first run the new check found TWO MORE that check-links.py could never have caught - one in direct-line.tsv carrying the central open question about Maria Lerena, and one written an hour earlier in sources-consulted.tsv - because check-links counts only PAIRS in rendered HTML, and a single unpaired marker leaves no pair to count.

how Adding the check after making the same mistake three times. The interesting part is what it found that the older guard structurally could not: a guard that looks for pairs is blind to the case of exactly one. And the new check then failed on THIS ROW, because the row originally quoted the markers as an example - the convention has no way to write a literal marker, which is now a known limit of it rather than a surprise.

was That Horace A Taylor married "(? Mary Jane Bridge)" - carried from a MyHeritage tree, hedged with the archive's own question mark

now HE IS HORACE ARCHIBALD TAYLOR and she is MARY JANE MARSHALL. Banns called at Northam, Devon, on 17 December 1893, when he was twenty-one - the earliest marriage of the ten siblings. The tree had the forenames right and the surname wrong, and the archive's question mark was warranted.

how Searching the Devon parish registers instead of trusting the tree. The same pass named five brothers-in-law this archive had recorded only as bare surnames - "m. Dalling", "m. Bell", "m. Burton", "m. Sweet", "m. Wood" - and it took under an hour.

was That the infant William Taylor (29 Dec 1824 - 10 Sep 1825) belonged to GENERATION A, alongside his own father James

now HE IS GENERATION B - James's son, and the elder brother of the William born in February 1826 who was named after him. The error had sat in data/taylor-line.tsv since the file was written.

how DRAWING THE CHART. Nobody had noticed in a fortnight of reading the file as a table, because a table does not care which row is which generation. The moment the same data was rendered as a tree, a dead infant appeared standing beside his father as though he were an uncle. A second way of looking at the same data is a test of it.

was Five entries on the places atlas that were not places at all - "SA records", "SA + Argentine records", "GRO + SA records", "Mendoza; later New York and Rio crossings", and one bare dash

now THEY WERE CITATIONS SITTING IN A PLACE FIELD, and they did TWO kinds of damage at once: eight people were missing from the map they belonged on, and four phantom locations were standing on it as though they were towns. Among the eight were PABLO ARMANDO HIMSELF, MARY SEPTIMA, and all three of their sons. The central figure of the archive was not on its own atlas. Now: Cape Town for the five South Africans, South Africa for Anton, Mendoza for Avelino Carlos, Montevideo for Carolina Oribe. The citations are kept in the notes, not deleted. 37 places became 33.

how David read the places data and spotted that the place field held a citation. consistency.py now flags any where_found or birthplace that reads like a citation - "records", "GRO", "crossings", "index", or an empty dash - and the check was verified by injecting one.

was A findmypast harvest plan listing OUTBOUND AND INBOUND PASSENGER LISTS as the second-highest priority, untried

now THEY WERE TRIED ON 9 SEPTEMBER 2026, under an earlier trial, and the results are in data/passenger-lists.tsv - six voyages with masters, ship numbers and TNA references, including the very record this session re-found and briefly believed new: Mary Septima's arrival at Southampton on the RMS Balmoral Castle, 20 May 1935, "C/O Dalling, Fore St, Northam". The eight-year hole had been worked too, with a clean negative and two forenameless candidates.

how Checking the archive's own files before writing up the "find". The re-find took five minutes; had it not been checked, a source consulted under a previous trial would have been announced as a discovery under this one. Read your own sources file before spending a deadline on a source.

was That HORACE ARCHIBALD TAYLOR "m. MARY JANE MARSHALL, banns 17 December 1893" - recorded this morning as a marriage, and used to correct a tree's "(? Mary Jane Bridge)"

now THE BANNS ARE REAL; THE MARRIAGE IS NOT EVIDENCED. Banns are an intention to marry. FreeBMD holds NO Horace Taylor marriage in volume 5b - the whole of Devon - between 1893 and 1895. He may have married elsewhere, later, or not to her. AND A DECOY NEARLY SETTLED IT THE WRONG WAY: a "Taylor, Horace" married a MARSHALL in the same quarter at BOSMERE, SUFFOLK (4a 1487), which would have neatly explained his absence from every Devon census - until the page was read directly and gave the bride as LAURA EMILY MARSHALL, not Mary Jane. Two Taylor-Marshall marriages, one quarter, two counties, two brides.

how Reading the volume and page directly instead of trusting the spouse-surname match. Before 1912 the GRO index does not print the spouse: FreeBMD infers it from a shared page, so a "match" is an inference and two couples of the same surnames in one quarter will produce a false one. FreeBMD is also explicitly incomplete, so the Devon null is softer than the others recorded today.

was Every verification run in this session, which took the form "python3 tools/consistency.py | tail -1" and was treated as a gate

now A PIPELINE EXITS WITH THE STATUS OF ITS LAST COMMAND - "tail" - WHICH ALWAYS SUCCEEDS. So consistency.py's exit code gated NOTHING for an entire day. The commits were chained with && behind that pipe, which means a hard contradiction could never have stopped one. It finally showed: an unpaired emphasis marker in errands.tsv was committed while the checker was printing CONTRADICTION on screen - the output was there and had simply scrolled past.

how Noticing that a commit succeeded in the same breath as a contradiction was printed. Fixed by tools/verify.sh: one script, set -e, no pipes, that runs all five gates and says ALL GATES PASSED or fails. The lesson is the archive's own, arriving in a new place: a guard that cannot fail is not a guard - and this time the guard was fine and the way it was CALLED was broken.

was Three rows written into errands.tsv with prose in the STATUS column - "DONE 14 September 2026", "PART READ - 57 of 141", and a whole paragraph - and one row written with three fields where the header has five

now STATUS IS A CONTROLLED VOCABULARY the errands page looks up: sent, replied, done, ready, waiting-on-david, blocked. Anything else falls through to "blocked", so three errands were being published to readers under the wrong chip - including one marked DONE that the page was calling blocked. And the malformed row put its whole detail paragraph into the status column and left detail empty. Now: David / waiting-on-david / the text in its proper field.

how tools/verify.sh, on its first real run. check-links caught the literal markers reaching the errands page, and chasing that exposed the column error behind it. The unpaired-marker check had passed the same row, because a field can be internally consistent and still be the WRONG FIELD.

was That the four LERINA arrivals of 1889 were excluded because "LERINA here is an Italian surname, not a variant of the Spanish Lerena"

now They ARE excluded, but on their dates and the composition of the party: they arrive in 1889 aged 59, 55, 21 and 17, seven years after Pablo Armando was born at Rosario, and the party contains no Juan Carlos and no Maria. The claim about the SURNAME is withdrawn as a reason - it was reached from an index without opening the manifest, which is the move that got the Armando Lorena card wrong.

how This archive's own search box, which showed all four LERINA entries carrying the same unexamined label.

was That the Arturo Lerena in army drawer Cajon 985 was born at Montevideo in 1858, son of Avelino Lerena and Ambrosia Traibel

now Two men. The Uruguayan record relied on gives the wife as EMA REGULES, and Ema Regules married JUAN ARTURO Lerena Traibel (1861-1936), whose children were born at Montevideo 1881-87. The Arturo in the drawer married AMALIA DIAZ and had children in La Pampa, Mendoza and Entre Rios 1889-97. The row is withdrawn and a "four branches" claim became three.

how Opening the FamilySearch Tree profile the record was attached to - going to the underlying record, which is the move this archive keeps having to relearn.

was Every person page showed an empty children list unless somebody had separately written an explicit "child" row

now A link is written once in data/relations.tsv - "X, father, Y" - and the builder recorded it on X's page and never reversed it onto Y's. Thirty-eight parent edges were read one way against fifteen compensating child rows. Pablo Armando, Juan Carlos, Maria Lerena, Mary Septima Taylor, Roque, Roseline Forbes and Doreen Chappell all had children in the data and none on their charts. Fixed in the builder, with aliases excluded so a machine-index variant is not counted as a second child.

how Wiring in the Montevideo household, whose four documented children rendered as none.

was Nothing, visibly - which is why it went unnoticed

now The people builder looked people up by name, and ELEVEN NAMES appeared twice or three times in the register - three Francisco Lerenas, three Jose Llerenas, two Carlos Lerenas, two Gilbertos. The later row silently won and thirteen rows could not be addressed at all. No edge happened to name an ambiguous person, so nothing looked wrong - but the next edge added would have attached to the wrong man in silence. Duplicates now carry a distinguishing tag and the builder REFUSES TO BUILD if two rows share a name.

how Caught on the way to adding twenty-four relationship edges, several of them to a "Carlos LERENA" and a "Gilberto LERENA".

was "Of the 1 relationships in the chart above"

now A plural where the count can be one. Three people pages carried it.

how An audit of all 138 person pages after the chart builder was changed.

was 129 people in the register

now 138, and climbing. The figure was typed by hand, correct when written and stale within days. It now derives from the data, like everything else on this site - the whole method being that pages are generated from the TSVs so page and data cannot drift. This one had escaped it.

how Reading the front page instead of assuming it.

was Paragraphs on six pages silently broke every emphasised phrase onto its own line

now The .prose rule is a CONTAINER style - display:flex, flex-direction:column, gap:16px - meant for a <div> wrapping several paragraphs. It had also been put on 21 individual <p> elements, and a flex container blockifies its children: every <em> and <strong> directly inside those paragraphs became a full-width block on its own line. About 27 phrases across /searched/, /argentina/, /taylor/, /register/, /method/ and /people/ read as broken-up fragments rather than sentences. Fixed with one rule, p.prose{display:block}, which keeps the intended 80ch measure.

how Looking at the page in a browser after adding a section, and not believing the explanation that it was my own markup - the CSS had no rule matching the element, so the cause had to be inherited from the parent.

was The archive emphasised things to its readers as like this

now The TSVs mark emphasis with triple asterisks and NOTHING EVER CONVERTED IT. 362 of them were rendering to readers across 38 pages - 132 on the register alone - so the archive was showing its own working notation as if it were text. Now converted at render time by site/src/lib/text.js, and check-links.py fails the build if a marker ever reaches a page again.

how Counting them, after noticing a pair of asterisks on the gaps page while fixing something else.

was The sources page said "Everything consulted so far" while listing nothing after 10 September

now It listed 12 sources and had not moved for three days, while the Montevideo household, the Rosario death index, 750 acts of the 1882 register, the full-text density measurement and Sancho's six spellings all went unrecorded. On a page whose entire purpose is letting a reader check the work, that heading was a false claim. It is now generated from data/sources-consulted.tsv - 33 sources, 13 of them PRIMARY - so it cannot fall behind again.

how Auditing my own list of what to do next, and finding the page contradicted the archive.

was That Luis Eugenio Lerena Lenguas and Julia Juanico had FIVE children

now SEVEN. Jose Julio, born 15 March 1879, and Candido Jaime, born 13 September 1884, were both missing. The five came from working outward from Julia's 1910 death certificate, which names five - and a death certificate is a list of the children who mattered to whoever registered the death, not a list of the children a woman bore. Found by searching the PARENTS' names directly instead.

how Doing the obvious search that had not been done: both parents' names, in one query, across every Uruguayan collection.

was That the Trinidad, Flores kindred included a daughter of Luis and Julia

now It did not, and that row linked the Montevideo trunk to the Trinidad kindred when nothing links them. "Luis x Julia, Emma Liseta 1876" was read off a findmypast list that gives mothers BY FIRST NAME ONLY, so the Montevideo couple looked like a Trinidad couple. She is Emma Sixta Lerena Juanico, born and baptised at Montevideo - our own trunk. The note's own prose said eleven Trinidad children while its table showed twelve; nobody had reconciled the two.

how Testing the register's search box with the term its own placeholder suggested - "Trinidad" - and getting nothing back.

was Ema Sixta Lerena Juanico born c.1880

now 21 November 1876. Three independent records agree: her civil birth registration of 30 May 1903, the Uruguay Baptisms index, and her christening at the Montevideo Catedral on 8 January 1881. The c.1880 was derived from her stated age at her 1913 marriage, and an age at marriage is the weakest date in genealogy.

how Searching both parents' names, which returned all three records side by side.

was The register promised "every named individual this archive has found"

now It omitted the entire Trinidad, Flores kindred - 29 people - although this archive had a whole note about them, calling them the founding kindred of the surname. The register's own search placeholder invited readers to try "Trinidad", which returned nothing at all. The people are now in it, and the placeholder is honest because the data caught up with the promise rather than the promise being weakened.

how Clicking the page's own suggested search term and watching it return zero.

was That Luis Eugenio Lerena Lenguas is not in the indexed Uruguayan death registers

now He is, and he died on 24 December 1901. Two sweeps were run and reported here as negatives - 1,502 Montevideo Lerena deaths for 1880-1911, and 132 Luis Lerena deaths for 1899-1903 - and this archive went so far as to write up the FamilySearch Tree's date of 1901 as a lead the registers appeared to contradict. The registers did not contradict it. THE INDEX DID: FamilySearch files him as "Luis SERENA Lenguas", so every search on the surname missed him. Found within the hour by searching his MOTHER, Justiniana Lenguas, instead.

how Searching on the rarest stable name in the household rather than on a surname that the clerks and the indexers between them spell Lerena, Lorena, Serena and Languas.

was Julia Juanico born about 1844

now 23 November 1842. The 1844 was derived from her stated age of 66 on her 1910 death registration; she was 67. Her baptism of 4 October 1843 at San Francisco de Asis gives the date, and gives her mother in full as Juliana JOAQUINA Gonzalez.

how Searching her FATHER rather than her surname - the same move that found her husband's death the same evening.

was That Bartolome Llerena was "the only person of this surname found in Rosario before 1882"

now Looser than the evidence allows. Six LORENAS are in the same 1869 census - Agustin, Guillermo, Geronima, Mercedes, Feliz and Maria - and this archive's own register groups LORENA with LERENA and LLERENA as spellings of one name, because the clerks disagreed. Two of them are flagged in that very register as the right age to be a parent of a child born in 1882. The note now claims only what was checked: the only person spelled Llerena or Lerena.

how A sweep of all 183 register rows against the notes that describe them - the note and the register contradicted each other.

was That FamilySearch film image arks cannot be reached by script, so each opening must be got by driving the viewer

now Wrong, and wrong because I tested the wrong URL form. /ark:/61903/3:1:<ARK>?i=<N> ignores the i parameter and keeps the same image - which is what I tested, and what I generalised from. /search/film/<DGS>?i=<N> RESOLVES TO IMAGE N, and the ark can be read straight out of the page title. I published the pessimistic claim in the register file and in a commit message within the hour.

how Trying the other URL form while starting the Rosario death registers, and getting a different ark back.

was That Luis Eugenio Lerena Lenguas was born about 1845, and his brother was "Alberto Otario"

now Born 9 FEBRUARY 1843, on the act of his own baptism - the 1845 came from his stated age of 56 on his death registration, and an age at death is an estimate. And his brother is ALBERTO OCTAVIO, not Alberto Otario; the latter is how FamilySearch indexed a hand that plainly reads Octavio. Their sister, indexed as "Justinia", is JUSTINIANA MAURICIA MAXIMA.

how Opening the 1846 and 1850 baptism images instead of trusting the index that summarised them.

was A row asserting that Maria E. C. Lerena, born 1869, was a daughter of Gilberto Lerena, born 1854

now He would have been FIFTEEN. All the record says is that she was in his household at the 1895 census; "daughter" was this archive's reading of a household, not the census's word. The row is left in place with the contradiction stated against it rather than quietly re-labelled.

how A new near-duplicate check in tools/consistency.py, which flagged her beside another Maria E. Lerena and made somebody read both rows.

was That "Armand[o] Lorena", born 1882 at Rosario de Lerma, Salta, was "probably not him" - written off on four objections

now It IS him. The image reads "Clase de 1882 / LERENA - Pablo Armando / hijo de JUAN CARLOS LERENA y de MARIA LERENA / Nacido en Rosario de Santa Fe / Oficina enroladora de CAPETOWN". Three of the four objections were artefacts of a machine index that dropped "Pablo", moved him 1,400 km, wrote Lerena as Lorena, and invented the surname "Serena" out of the printed word SERIE.

how Opening the image. It took two minutes. This is the most serious error in the archive: the central finding of the project was nearly discarded.

was That Roque's first wife was "Rosalina Wilhelmina CHAPPELL", read from a marriage index

now CHAPPELL was her MARRIED name. She was Roseline Wilhelmina FORBES, widow of Frederick Christian Chappell. The 1936 register calls her "Rosaline Wilhermina Forbes Chappell" because that is what a remarrying widow is called.

how The 1920 Chappell-Forbes marriage, found in BOTH the civil and the Anglican registers, plus Chappell's own estate file.

was Margin readings taken from register openings rendered at 42% scale

now Not reliable. At that size "Timoteo Lucero" was read as "Armando Sara" - both words wrong, and an L-surname missed, which is the exact failure that matters when hunting one L-surname. Every opening was re-read from the margin strips at maximum resolution.

how Zooming on a single doubtful entry, and finding it was not doubtful at all - it was simply wrong.

was That the Lerena palette was identical to the FALCO archive's, with hex values quoted

now The values quoted were Falco's, not ours. The Lerena styles.css is byte-identical to the DEFRANCESKI archive's. Falco is the outlier, with a warm palette against the shared cool one.

how Reading both files side by side instead of trusting memory.

was That there were 29 places in the register

now 28. "arr. Buenos Aires 1907-07-14" had been sitting as its own one-person place beside the 43-person "Buenos Aires - arrivals" because the normalising pattern only matched "arr. YYYY".

how Looking at a person page that had no business having its own place.

was Layout: a bare "6" floating in the middle of a sentence on every person page

now .prose is a flex COLUMN, not a paragraph class. Using it on a <p> made every inline <strong> its own row.

how Looking at the rendered page rather than the built HTML.

was That the Italian-origin hypothesis was "the most consequential thing on this page"

now Within the hour, the low-confidence genetic groups showed no Italian group among the five legible ones - evidence against it. Downgraded the same day to "already weaker than when it was written".

how The next screenshot.

was That the Rosario marriage index showed "Larena V x Sarmiento A" at Libro 9 folio 20 - and that this was a candidate

now The entry reads LAVENA V. x BELMONTE A.. It was misread at low resolution and kept a false candidate alive for two days.

how Read again at the film's maximum resolution on 10 September. Not close: the hand is unambiguous.

was That a cluster of Larena/Lerena at Cayasta might be a kindred

now Cayasta is an ITALIAN colony - Laurini five times, Lauron, Castelarin - and the same mother is transcribed Azela in one act and Raiela in another. The "cluster" was a transcription artefact.

how Reading the acts instead of the index. Retracted publicly; the register entry is now EXCLUDED.

was That the 1869 census showed a Llerena HOUSEHOLD at Rosario

now The image shows twelve separately-numbered, unrelated individuals in a conventillo. There is ONE man of the name, Bartolome, not a family.

how Opening the image instead of trusting the index grouping.

was That Gabino L. Llerena was a Rosario-born link between the two countries

now Every other person on his record is Uruguayan, so the Rosario in question is Rosario, COLONIA, in Uruguay.

how Caught before it was published, by reading the rest of the record.

was That reading the Rosario registers needed the Chrome extension, because the in-app browser could not crop

now Wrong twice. The viewer has its own zoom, and - far more important - the Image Index panel is an HTML TABLE. The handwriting never needed to be read at all.

how David: "why do you need chrome? use your own browser?"

was That CEMLA lived at cemla.org

now cemla.org is a Mexican central-banking body. The arrivals index is at cemla.com/buscador/

how David went to the address given and found the wrong institution.

was That Mary Ann Hellen Brayley was a new find

now She had been in data/taylor-line.tsv since 8 September, from David's own MyHeritage tree. What findmypast supplied was a citation - the baptism of 30 Oct 1884 at Northam, ref 1843A/PR/1/16 - not the fact.

how David: "why is her mothers name new? its been in my myheritage tree forever."