Behind the paywall
A marriage this archive had never found, and four generations it had called untestable. All of it was one subscription away, and none of it was ever missing.
Two of this archive's oldest walls came down in an afternoon, and neither was a wall. The coverage page had measured every free index honestly and concluded that certain records could not be reached. They could be reached. They cost money.
Robert D'Arcy married Jean Ward
ROBERT DARCY married JEAN WARD at PORTSEA ST MARY on 21 JUNE 1779.
Joseph D'Arcy, baptised Portsea St Mary 19 March 1780, SON, father ROBERT, mother JEAN — the Hampshire Genealogical Society's own transcription, independent of the FreeREG entry this archive has been quoting.
Put beside what the archive already holds, the shape of a life appears: married at Portsea in June 1779, a son christened there in March 1780, a daughter married at Chatham in 1806 with that son signing as witness, a colonelcy gazetted in 1813, and a burial at Chatham in May 1827 aged seventy-six. Chatham → · The Hornby claim →
Joseph had no brothers or sisters at Portsea
Test: John Smith at Portsea, 1770–1794, in FindMyPast's Hampshire collections. Result: Portsea St Mary baptisms come back for 1770, 1771, 1772, 1774, 1775, 1779, 1782 (twice), 1785, 1786, 1788 (twice), 1789, 1790, 1791 and 1792 — continuous across the whole window, where FreeREG holds twenty-two months.
Which is not the same as saying they had one child. A regiment moves. The couple married at Portsea in 1779, baptised Joseph there in 1780, and by 1806 a daughter was marrying at Chatham — the children in between will be in garrison parishes this search has not touched.a parish, not a life
The page that said this could not be tested →
The generations that were never untestable
The archive has said for months that five of the tree's six Saniger generations could not be tested because FreeREG's Berkeley transcript stops in 1677. Every one of the following is a Berkeley record from after 1677.
| Who | Event | Date | Parish | What it names |
|---|---|---|---|---|
| Thomas Saniger The tree gives 23 August 1744 at Berkeley. The same day. | baptised | 23 Aug 1744 | Berkeley | father THOMAS Saniger |
| Thomas Saniger A name this archive has never had. John was baptised five weeks later. | married | 10 Apr 1769 | Berkeley | to CATHERINE COTTON |
| Jn'o Saniger The tree gives John's birth as 15 May 1769. Baptised three days later. | baptised | 18 May 1769 | Berkeley | father THO'S Saniger |
| William Saniger Hannah's brother. | baptised | 6 Sep 1801 | Berkeley | father JOHN Saniger |
| Jno Saniger Hannah's brother. | baptised | 18 Sep 1803 | Berkeley | father JNO Saniger |
| Thos Saniger Hannah's brother. | baptised | 3 Jul 1808 | Berkeley | father JOHN Saniger |
| Thomas Saniger Born about 1744. The tree says he died about 1819. | buried | 24 Jan 1819 | Berkeley St Mary the Virgin | AGED 75 |
| John Saniger Born about 1769. The tree says he died on 10 March 1824 — ten months out. | buried | 16 Jan 1825 | Berkeley St Mary the Virgin | AGED 55 |
Hannah herself is still missing from the index. Hannah's own baptism is not in the index. Her father's children appear at Berkeley in 1801, 1803 and 1808, and the tree puts her birth on 23 July 1805 — exactly in the space between. That is a gap in a transcription, not an absence in a register: FamilySearch's England Births & Baptisms is not a complete copy of Berkeley St Mary, and the original book is what will settle it.
Two things the tree had wrong
| The tree says | The register says |
|---|---|
| John Saniger died 10 March 1824 at Berkeley | Buried at Berkeley St Mary the Virgin on 16 JANUARY 1825, aged 55. The tree is ten months early. |
| Hannah Saniger was born at Berkeley on 23 July 1805 | Not yet found, but her father's other children are baptised there in 1801, 1803 and 1808, and the 1851 census gives her birthplace as Berkeley. |
What this says about everything else here
This archive has been careful about coverage and careless about one word. It measured what the free indexes hold, correctly, and then let not reachable from here become not there. Those are different sentences and only one of them is about the past.
The will of a Chepstow mariner, digitised and unreadable
Everything above is a wall made of money, and money is the easiest kind to describe: it has a price, and paying it ends the argument. This one has no price.
EDWARD SANNIGER, mariner of Chepstow, died 1806, and his will survives twice over. IR 26/352/152 at Kew is the legacy-duty abstract; LL/1806/24 at the National Library of Wales is the will itself. The reference is firm, the catalogue describes it, and the document is digitised. It is also unreadable from here, because archives.library.wales serves a bot check — «Performing security verification… protect against malicious bots» — before it serves the image.
Every marriage this archive cites, re-opened
On 16 September a claim carried by every online tree — that John Saniger married Lydia Jones at Berkeley in 1797 — was disproved by opening a transcript and reading one field the result row does not show: the spouse's surname. The man was a Sawyer. That raised an obvious question about this archive's own work, and it has now been answered rather than assumed.
| The claim | Set | Spouse's surname on the transcript |
|---|---|---|
| Robert Darcy m. Jean Ward · Portsea St Mary, 21 June 1779 | Hampshire Marriages | WARD — confirmed |
| Hannah Saniger m. James Wakefield · Bristol, 17 October 1831 | England Marriages 1538-1973 | WAKEFIELD — confirmed |
| Thomas Saniger m. Catherine Cotton · Berkeley, 10 April 1769 | England Marriages 1538-1973 | COTTON — confirmed |
| Thomas Sanigre m. Martha Golding · Bristol, 24 August 1828 | England Marriages 1538-1973 | GOLDING — confirmed |
| Joseph Ward m. Marth Gairden · South Leith, 27 October 1747 | Scotland, Parish Marriages & Banns 1561-1893 | GAIRDEN — confirmed, and her father with her |
documented
datasetname tried. It resolves through a cross-set search with a place instead — place=south+leith and no dataset at all. The transcript confirms GAIRDEN, and carries a field the result row never shows: Spouse's father — WILLIAM GAIRDEN. The argument made for it on the 17th, that a result row has never carried a father so somebody must have read past the row, turns out to have been right. It is now checked rather than argued. All five hold. documented9 ways this site returns a number that is not about the world
A null is only a fact about the world if the search could have found the thing it did not find — and a total is only a fact about the world if the thing it counts is the thing it is called. Over 20 and 21 September this archive caught FindMyPast doing both wrong, in the ways listed below, each of which had already been published here as a finding or was about to be. They are written down together because the next session will meet them again.
county — it cost: the sisters, 1861CATHERINE SMITH born about 1789 in Sussex returns nothing in the 1861 census, which cannot be true of any English county. The combination silently returns zero on the census sets. Three searches for the Brighton sisters had been built on it.
yearofrecord on the Army Lists — it cost: Robert D'Arcy's «blank»Asking for SMITH in 1790 returns twenty rows dated 1673, 1678, 1689, 1690, 1692, 1693, 1704, 1707, 1709, 1781, 1802, 1803 and 1826 — and no 1790 at all. The parameter is thrown away and the search answers a different question.
The identical query for COTTON baptisms in Gloucestershire returns nothing sorted ascending and a page of results sorted descending. Results render asynchronously, so a zero read too soon is indistinguishable from a real zero — and nothing on the page says which it was.
yearofbaptism against an entry indexed by birth — it cost: the 1770 RobertEngland Births & Baptisms holds two kinds of entry — some with a baptism year, some with only a birth year — and the filters are not interchangeable. DARCY unfiltered returns rows dated 1666, 1707 and 1708 whose «Baptism year» column is empty, so every yearofbaptism search against that surname returns nothing whatever the years.
SANIGRE returns Thomas Sinegar's 1828 marriage and nothing else in England returns it. SANIGER asked the same way in the same county and window returns 8 marriages, SANIGAR returns 3, and he is in neither. The surname box answers for a string, not for a name — so a null recorded as «not found in any spelling» is only ever as wide as the list of spellings somebody typed, and the list here came from a baptism index in which this form does not occur at all.
298 index entries across the five commonest spellings of this surname are 246 baptisms. The set transcribes registers and bishops' transcripts and some parishes twice over: Mary, daughter of Samuell, is indexed four times at Berkeley in 1688, and Sarah, daughter of William, four times at Bristol in 1839. The duplication is not even — SANIGER and SANIGAR hold 48 of the 50 repeats and SINNEGAR holds none — so an entry count measures how often a parish was typed up as much as how often a name was written.
The GRO death index prints a District column and is addressed by district. Give it place or registrationdistrict instead — both perfectly reasonable names, one of them the column's own full title — and it returns nought with no complaint. SMITH at place=lymington returns zero, which cannot be true of any English district; district=reading narrows 128 Joseph Smiths of 1848 down to the right one. Nothing on the page distinguishes a parameter it ignored from a record that is not there.
«Newspapers & Publications» here is a name index to BMD notices, not the British Newspaper Archive's full text, and its Year column is blank for most rows. SMITH returns 674,001 notices unfiltered and 0 with a year of death of 1827 ± 2. The filter is working exactly as asked and discarding everything that was never dated. Forename filtering does work — Robert D'Arcy narrows 4,459 rows to 42 — so nothing on the page suggests the year is the problem.
yearofbaptism_offset accepts any integer, not only the seven in the dropdown — SANIGER baptised 1792 returns 5 at offset 0, 7 at offset 3 and 58 at offset 40. Above 40 it does not widen, and it does not complain: offset 45 returns 5, which is exactly the offset-0 count. The widest search a reader can ask for silently becomes the narrowest one there is. This is the worst of the faults on this page, because every other one hands back a zero and a zero at least looks like an answer worth checking. This hands back a real, small, plausible result with nothing wrong with its face. Measured here on 21 September after the Defranceski session found it on an Italian surname and corrected its own first account of it.
documented
And on 21 September the rule gained a method, not just a warning. The surname box takes wildcards. S*N*G* asked of Gloucestershire marriages for 1827-1829 returns 53 rows, and among them is Thomas Sanigre, 1828 — the marriage eight named spellings had failed to find and this archive had recorded as absent — together with one marriage indexed as both SANIGER and SANIGOR at Rodborough in 1827. A list of spellings is a guess about a clerk; a wildcard is not. The cost is strangers — the same 53 rows hold Sanger, Shearing and Sheering — so the trade is a wider net inside a narrow window, read row by row, rather than a narrow net cast across a century. Where a question has a place and a decade, that is now the first search and the spelling list is the second.
And the same day showed what «the same query shape» does not cover. This archive published «None, in any spelling» for Sanigers at Chew Magna with a control beside it, and the parish register holds two — Charles Sinagar 1816 and William Sinegar 1822, children of Simon and Elizabeth. The control was a search at Berkeley. It returned 252 rows, so it proved the site was answering; it could not prove that the Chew Magna page had finished rendering, which is the one thing that had to be true. A control that changes the field under test is not testing that field. Where a query takes a wildcard, the wildcard is the better control: it re-asks the same field in a way that cannot fail for the reason a spelling list fails. S*N*G*R at Chew Magna returns four records and two of them are the name. check_controls.py now refuses a new FindMyPast null that claims «any spelling» without one.
One line added to that on the evening of the 21st, from the Defranceski session, and it is the operational half. A wildcard buys reach by giving up discrimination, and it does not report which of the two it is currently doing. So make it report: run the exact spelling first and read the ratio against the wildcard. One in twenty here; one in a hundred and twenty for a surname whose stem is native English. That single extra query is the difference between a tool that saves a day and one that invents a family, and it is cheaper than either.
documented21 September added two more, and both had already been published. The spelling table's headline said «371 baptisms» where it had counted 371 index entries, on two pages. And a null saying Thomas Sinegar's marriage was «not found, in any spelling» had stood on /searched for eight days while /thomas-sinegar carried the marriage itself. Own errors 38 and 39.
One piece of good news, and it is structural. A datasetname that names no real set returns a 500 error, not a zero — proved by asking for a non-conformist set that does not exist beside one that does. So a zero from this site is always a zero from a set that is really there. It does nothing about the faults above, where the set is real and the filter is the liar, but it closes off a whole class of false null that a mistyped set name would otherwise create silently.
Sources: FindMyPast — Hampshire Baptisms and Marriages (Hampshire Genealogical Society); England Births & Baptisms and England Marriages 1538–1975; National Burial Index (Gloucestershire FHS)