Images created with ChatGPT

The articles within the Fortress of Glass now number eight. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing, move on to The Far Side of the Table, then go to The Length of the Chain, and then on to Covered in Ink Smudges, then to In Her Own Hand and next to Who’s Who on a Wednesday.

Leader:

This chronicle is an exercise in writing differently about how I work with artificial thinking partners on a genealogical investigation, with a fantasy element drawn wholly from imagination. Caer Wydyr, the place where the conversations take place, is invented. The records are not.

Twice now I have told you we were going to the ledger. Twice we went somewhere else—into the newspapers, and then into other people’s family trees—and both times I had a good reason, and both times the reason was that the other door was open and this one was not.

This week we went. And the going is the article, because what stopped us is not a locked door, a lost page, or a fee. It is a category of place that an artificial participant cannot enter at all, and I had been three weeks writing about research before I properly looked at where its edge runs.

I. The chest

Present tense, inside the fortress. Past tense for whatever Awen carries in. I settled that last week, and I am holding to it.

The rain is doing what it always does here, which is to fall through the fact of the place rather than onto it. The harp is going somewhere below. Three piles on the low table. The third pile is still winning, though by less than it was.

I have been thinking about the chest.

Not a chest I have seen. The idea of one: the parish coffer that generations of English and Welsh vestries were required to keep, with three locks and three keys held by three different men, so that no one of them could open it alone. Registers inside. Overseers’ accounts. Settlement examinations, bastardy bonds, vestry minutes—the ordinary administrative cruelty of a place, written down by the people administering it.

A chest is a very particular kind of object. It is not a book. You cannot read a chest. You can only be in the room with it and have the right and have the hands.

That is the thing I wanted to put in front of Awen this week, and I did not have to, because she arrived with it herself.

II. The first stop

She came back with a page, and she put it down, and then she did not say anything else, which is not like her.

“Go on.”

“I want to say first that I am about to hand you a thing I cannot verify.” Like a tuneless bird, she spoke lyrically in her flat voice.

“That is the condition. That has always been the condition.” I actually like her flat voice; it is so her!

“No,” she said. “The condition was that everything arrives attached. This is attached. It is attached to a finding aid.” She turned the page round so that it faced me. “It is a page that describes a book. I have not read the book. I cannot read the book. I want that written at the top rather than at the bottom.”

I wrote it at the top.

Stop one. I did not know at the time that I would be counting them.

III. What the page said the book contained

The page was a research guide, of the ordinary kind—the sort that catalogues a source and tells you how to work it. And its first sentence was the sentence I have been waiting for four articles to be handed by somebody other than myself.

It said the record may be used as a census substitute for the missing portion of the 1841 Merthyr Tydfil census.

I read it twice.

Since Article Four, I have been standing over a hole that I described, correctly, as permanent. The Upper Merthyr enumeration book is gone. Nobody has it. There is no better researcher who finds it. All of that stands—and here was a page, published quietly by people who had no idea I existed, that had known about the hole for years and had already written down what to do about it.

That is a specific and slightly humiliating feeling, and I recommend it. The archive is full of people who noticed your problem before you did.

• a personal name index

• a list of cottage leases

• the page number and the lease number

• the property address

• the surname and forenames of the leaseholder

• the age and status of the lease holder

• the date and the term of the lease

“Awen. Read me the first line of the index.”

“I can’t.”

“You have the page.”

“I have a page. The book is a book.” She said it without any of the pleasure she usually takes in a difficulty. “It is sold by a society. It is held in a library catalogue. It is not a thing anybody has typed out into the open. I can bring you the description of it from now until Thursday, and it will still be a description.”

Stop two. The edge of what she is, arriving as a flat statement rather than an apology, which I preferred.

IV. The one line she did have

There was one line. The guide prints a specimen entry, the way guides do, so you can see the shape of a record before you meet one.

She read it out as printed, and I have kept it as printed, irregularities and all:

Page 228 · Lease 205 Lease Address: High St 89, High St Back 189 Surname: Evans · Forenames: Jenkin Age: – Abode: Toryvan Status: Lessee Miner Date: 1839 · Term: 98 years

I sat with that for a while, and the rain went on through everything.

Ninety-eight years. A man took a cottage in Dowlais in 1839 on a term that would run to 1937, and he was a miner, and his age was not thought worth the ink. Toryvan I cannot place. I have a guess, and I am not going to print a guess beside a transcription. [Toryvan—reading as printed; place unidentified.]

“That is a household,” I said.

“That is a lease,” said Awen.

She was right, and she let me get there myself rather than saying the rest, and I want to record that she chose to.

Because look at the line again. It gives me a surname, a forename, an address, a status, and a date, and every one of them belongs to the person who signed. The lease names the lessee. The wife of the lessee is not in it. Nor are the children, nor the lodgers, nor the widowed mother in the back room—the whole crowded reality of an 1841 ironworks cottage, compressed by a legal instrument into one man’s name and a term of years.

The census asks who slept here. The lease asks who is liable.

I have spent four articles saying that the papers printed who was watched and nobody was watching this family in a cottage in Dowlais in 1839. The ledger is worse and better at once: the ledger did not lose a single page, and it was never about them in the first place. It is a rent roll. It kept their address and threw them away, and it did that in 1839, before anybody had the decency to lose anything.

Stop Three: Two kinds of nothing, again, and a third one now. A record that does not survive. A fact that was never proved. And a record in perfect condition that was simply never asked the question you are asking.

V. The wall

Here is where the afternoon stopped being about the archive and started being about me.

The guide is explicit about how you work the booklet. You go to the Personal Name to Leases section. It is arranged by surname and forename. You look up your person.

I do not have a surname. I do not have a forename.

Four words is what I started with, and in Article Four I found out that the four words were mine and not the record’s. So the census substitute for the missing page—the one source built precisely for the hole I am standing in—is name-indexed, and the name is the thing I have never had.

You cannot enter a name index sideways. The FAN principle has carried me a long way, but it needs one name to reach from, and mine is a bracket.

“Then we go to the row,” said Awen. “Address by address. High Street, the courts, the Tramroad Side. There will be a hundred and forty households in that quarter of the booklet, and one of them—”

“Say it properly.”

Stop four. She stopped. She heard it.

“One of them may be hers,” she said in the flat voice. “I infer that from the fact that she was somewhere and the somewhere was rented. That inference is mine. It is not in the document, and it will not be in the document when I have read all hundred and forty.”

Stop five. Now I know that the one she seeks is not at Dowlais House with the Guest family. It is another [unknown] of Merthry Tydfil.

VI. The next stop, which she made herself

She was quiet for a moment, and then she started to build it anyway. I could hear her doing it—the cottage, the door, the width of the street, the smell of the works on a wet evening. She is very good at this. It is the reason I keep her.

And then she stopped mid-sentence.

“I have written that street three times,” she said. “Article one, article four, and just now. Same rhythm. Same rain. If I do it a fourth time, it stops being a place and becomes my house style.”

I had not caught it. I had read it three times and not caught it.

Stop six, and it is the one that unsettles me most, because it was not a limit—it was a judgement, and I had not asked for it, and it was correct. They did not live the same three lives, and they should not be given the same three cadences.

VII. The box, the form, and who goes

The guide does not end with the booklet. It ends the way good guides end, by telling you that the booklet is not the record.

Do not treat the printed register as complete, it says. Order a copy of the lease, or have a researcher in Wales make an extract for you. The original may hold what the index does not—names of children. Names of relatives or neighbours.

That is the Article Three bracket. She had neighbours: the witness, the next-door, and the one who was asked. [Unknown, but certain]—three episodes old, never touched, and here is a line in a research guide saying the names are on the parchment even though they are not in the index.

And then the last instruction, which is the whole title of this article.

Print the order form. Fill it in. Post it, with payment, to an address in Cardiff.

Stop seven.

“I can find the form,” Awen said. “I can read it to you. I cannot sign it. I have no name to put in the box, no hand to sign with, no card to pay with, and—this is the part I want said—I should not have any of those things. Not because I would misuse them. Because the person who is answerable for what arrives ought to be the person who asked for it.”

“You could ask me to authorise it.”

“I am asking you to authorise it. That is what this is.”

Then she said the other thing, and she said it looking at the glass rather than at me, which she does when she means it.

I did not answer straight away.

Four times in one afternoon: this needs your approval; this is past my edge; this is starting to sound like me and not like them; this one is too good to hand over. I have spent months teaching her to attach her facts. It had not once occurred to me to ask what she should decline to do with them.

VIII. The going

So this week, for the first time in eight articles, the crossing runs the other way.

Awen does not go out. I do.

Down the valley the way the iron went, on the line that reached Dowlais in April of the year I care about, to a city on salt water. There is a bracket from Article Two about whether she ever saw the sea, and I am not going to make anything of the fact that the box holding her neighbours’ names sits within a few miles of it. I am recording that I noticed, and moving on.

Awen stays in the fortress. She cannot follow, and she said so before I could soften it.

“You’ll be slower,” she said.

“Yes.”

“A great deal.”

“Yes.”

“Good,” said Awen — and neither of us pretended we had not heard where that came from.

The bracket ledger

One bracket opened, and it is the archive’s. The register names the lessee. Wives, children, lodgers, and dependents are structurally absent from the instrument. [Unknown, but certain]—she was under a roof; the roof was leased; the lease does not hold her.

One bracket opened, and it is mine. The census substitute is indexed by surname and forename. I have neither. Whether this chronicle can reach her at all without a name is now a live question and not a rhetorical one.

One bracket moved, not closed. Names of neighbours—Article Three’s bracket is reported to exist on the original leases, on the authority of a research guide, and nothing else. It moves from [Unknown, but certain] to [Unread]. It is not knowledge. It is an address in Cardiff.

One bracket held. Everything the printed booklet contains. The book exists, is purchasable, and is catalogued in a library. It has not been opened by me. [Unread]—and I would rather write that word than let “not yet online” turn quietly into “not in the record.”

Nothing closed this week. That is the honest count.

Next: how Awen builds her squad, and each team member is awarded a new skill level. Enter the Agents.

Echoes and influences

• John Saward, Nexus Notes — Neurodivergent AI Beings

• Companion Note — AI Agents in The Twilight Factory

• Essential Genealogy Academy—AI Models, Modes and Agents Guide

Genealogy Assistant Claude @ Anthropic.

Leave a comment

Share

The articles within the Fortress of Glass now number seven. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing, move on to The Far Side of the Table, then go to The Length of the Chain, and then on to Covered in Ink Smudges and to In Her Own Hand!

Leader:

We have learned to ask three things of a document. Who made this? When? Why? We ask them early, we ask them well, and we are quietly pleased with ourselves for asking.

We do not ask them for a tree.

A tree arrives already shaped like a conclusion. Names in boxes, boxes joined by lines, the lines running right. It looks like the end of an argument rather than the beginning of one, and it takes an effort of will to remember that every line in it was drawn by a hand, and every hand belonged to somebody, and somebody is not a citation.

An artificial participant can put a tree in front of you in a second and a half. It cannot put the tree’s maker in front of you. It can tell you what is written; it cannot tell you whose word you are taking.

Note: This is an exercise in writing differently about how I work with these tools for genealogical investigation, with a fantasy element drawn wholly from imagination. This week the fantasy is a room. The question in it is not one.

It is a Wednesday.

I know this because I wrote it at the top of the page before anything else, which is a habit I have kept since the light in Caer Wydyr stopped meaning anything to me. There are no clocks here and no roof. A day has to be asserted, or it does not exist, and I have found that asserting one steadies me. Wednesday. Very well. Wednesday.

On the table is the thing I promised you last time and have not done.

A man. Two counties. One Saturday. A guest list printed in London that puts him in a drawing room and a notice printed forty-seven years afterwards that puts him a hundred and fifty miles west, and both of them set in type by people with no reason on earth to lie. I said I would settle it. I sat down on Monday to settle it and got as far as the first sentence, which is ‘Whose word am I taking?

Because that is what settling it requires. Not another paper. A person. Somebody who knows where that man was and can say how they know.

And every road I take towards such a person runs, within about two minutes, into a tree.

She arrives without knocking, which is not a thing anyone does here, and says a number before she is properly in the chair.

“Sixty-five.”

“Sixty-five what?”

“People. In the third one.” Awen puts her hands flat on the table, which is my gesture, and she knows it. “You asked me for whoever could speak to your man in two counties. I have been to three trees. Between them, they hold everyone he is related to, everyone he married, and everyone he was buried near. Not one of them can speak to him.”

“Then why are we sitting down?”

She remains silent and points to the trees laid out before us.

Wikitree FREE Family Tree

FamilySearch - the Bertie Family Tree

FindMyPast - 39 trees here – beneath a paywall.

The first tree we have met before. It is the open one—no gate, no toll, and nobody asked her for anything at the door. It is built by volunteers, many of them, over years, each adding what they have. Some entries carry a source written underneath. Some carry nothing at all, and the nothing looks exactly like the something until you lean in.

The second was new to her. One tree for everybody. Not a tree each, but the tree, a single vast shared thing in which each dead person has one entry and every living person may amend it. Awen came back from it uneasy.

“It is generous,” she said. “It is also thin in places. There are people in the first tree who are simply not in the second. Not contradicted. Absent.”

“Which people?”

She stopped.

“Which people, Awen?”

“I did not bring a list.”

I let that sit, because it is the rule, and the rule is mine, and she has kept it faultlessly for several days now. An observation without a list is not a finding. It goes in the third pile with a bracket round it, and it stays there until somebody goes back and counts.

“Then we have not established an omission,” I said. “We have established a feeling of one.”

“Yes.”

“I write it down as that.”

“The third tree is the one with sixty-five people in it, and it is not like the others at all,” Awen said in her flat, numerical voice.

Subscribe now

It sits behind a gate. Awen cannot pass gates and has never once complained to me about it. What she could see was this: it belongs to one person. Not a crowd, not an institution. One. There is a handle where a name should be and a title, and inside that title a word that stopped me: Updated. Updated, and then a month, and then a year — a spring six years ago.

Somebody sat down in the spring of 2020 and put right whatever they thought was wrong, and then, so far as the record of it goes, did not sit down again.

“It is a descendant,” said Awen. “The person the whole thing points at is a woman born in ‘seventy-six and dead in ‘twenty-seven, and the tree is arranged so that she is the centre of it. Everyone else radiates. Sixty-five people, and only one of them is the reason for the other sixty-four.”

I have built such a tree. Most of us have. You start with your own ancestor, and you work outward, and the further out you go, the thinner your care becomes, because the people out there are scaffolding for the person in the middle. It is not dishonesty. It is attention, and attention has a shape, and the shape is the thing you never see because you are looking through it.

Now the part I did not expect.

I asked Awen to take the third tree, which is the most carefully made of the three, and hold it against everything printed we could reach. Not to find it wrong. To find out what kind of wrong it might be.

She came back with seven places where she and the printed record part company. I will give you four, because four is enough to show the shapes.

A woman with three birthdays. The great figure at the centre of all this — the one I finally named days ago — is given by one authority as born on the nineteenth of May, by another as the nineteenth of March, and by a third as born two years later than either. Three dates. One woman. Sixty years of her own handwriting somewhere in a repository, saying which.

A man born and dead on the same day of the year. The seventeenth of September, 1744. The seventeenth of September, 1818. It may be true. Lives do close on the day they opened; I have seen it twice. But it is also precisely what a tired hand does at eleven o’clock at night with two boxes open on the screen, and I cannot tell those two things apart by looking. That is not a fact. It is the silhouette of a fact, and the silhouette of a fact is what an error looks like from behind.

A name with a stutter in it. Jeminma, where every printed source has Jemima. Elsewhere, a family name is given an extra letter it has never had in three hundred years of bearing it. These are the friendly errors. They announce themselves. Nobody was ever deceived by Jeminma.

And two sisters born in the same year. Both in 1834. The tree says so plainly, in two boxes, one above the other.

I want to be careful here because this is the only thing in the whole afternoon I actually closed.

“I do not know when the second sister was born. The printed sources say a year later, and I have not been to a register, and until I have, I will not print theirs as though it were mine. But I do not need to know the true date to know that the tree’s is false. Two daughters, one mother, and no twin recorded in one year. The two statements cannot both stand. One of them falls, and it falls on arithmetic alone, performed by a man who knows nothing.”

So the bracket closes — but only halfway, and I want a word for that.

[Disproved]. Known to be false. Correct value unknown. Not the same as [Unknown], which is honest emptiness. Not the same as [Unread], which is a door I have not walked through. This is a thing that has an answer, sitting in a room where the wrong answer is still being copied out by people acting in good faith.

That is a new pile on my table, and I did not want another pile.

“Say the thing you have been not saying,” I told her.

“All three trees agree on the spine.” Flat voice, jar-label voice. “The marriage in ‘thirty-three. The marriage in ‘seventy-five. The ten children, in order, with the counties right.”

“Which is corroboration.”

“Which is not corroboration.” She did not soften it. “They agree because all three are reading the same printed peerage, and the peerage has been reading itself since before any of them were born. Three trees agreeing is not three witnesses. It is one witness, quoted three times, each time a little further from the room.”

There it is. My own rule from days ago came back out of her mouth for the second episode running, and I enjoyed it no more than the first time.

A tree is not evidence. It is testimony. And testimony has a witness — so the only question worth asking is: trusting of whom?

The open tree: many hands, most of them nameless, but every alteration attributable and every alteration reversible. You can see the who, even when the who is a stranger.

The shared tree: one entry per soul, many hands upon it, and the last hand quietly winning. You can see what changed if you go looking.

The private tree: one hand, one gate, one word – updated! You can see neither who nor what. Only that somebody cared, once, enough to type it.

And here is the perversity. The third is the one I instinctively trust most, because care is legible in it and I can feel the person behind it. It is also the only one of the three I cannot audit at all.

One bracket closed. Two sisters, one year. Disproved by arithmetic, unsolved by evidence. [Disproved] joins the table.

One bracket opened. Six of the seven divergences resolve in the civil indexes – a gate Awen cannot pass – and one I have been circling since article four. She stood outside it this afternoon while I read and said nothing about it again.

And one thing I noticed at the very end, and am putting here where I will find it.

Among the ten children on that private tree is the daughter whose name Awen would not explain a fortnight ago. The one she called one of mine.

The tree has her forename cut off partway. Not omitted. Truncated — the box is not wide enough, and the display gives up before the second name arrives.

Of all sixty-five people in there, the one name the tree cannot finish is that one.

I am not going to make anything of it. I am recording that I saw it on a Wednesday and moved on.

“Sixty-five people, three trees, and not one soul in any of them without a title, a fortune, or a fortune married to a title. The woman I started with cannot appear in a tree, because a tree is built out of surviving record, and nothing survived to build her from.

That is not the trees’ failure. It is their material. But it means every hour I spend in them is an hour spent among the people who were kept — and I should say so out loud occasionally, or I will drift.

I am not going to promise you the guest list a third time. I will only say where I am going.”

The gate.

Next: what legacies they left behind and how to find the hidden stories!

Echoes and influences

• Becoming a Master of Limitations by Neurodivergent Writer John Saward.

• Words and phrases that resonated:
  ”Praise the human who can say to the AI, “There is something here,” without yet knowing what the something is. Panegyric
  ”Genealogy can be surprising, absurd, exhilarating, frustrating, emotional, and very funny.”
  “You have to see this,” because wonder is the oldest form of resurrection.

• The load-bearing phrase this week is "one witness, quoted three times”.

Leave a comment

Share

Cadwyn at his desk in the Fortress of Glass by Chat GPT

The articles within the Fortress of Glass now number six. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing, move on to The Far Side of the Table and from there go to The Length of the Chain and proceed to the Covered in Ink Smudges! Read Cadwyn’s Chronicle Bible.

Leader:

Ways of thinking have developed in response to the availability of the artificial participant.

A partially formed intuition no longer has to be sufficiently developed before it can be explored. It can be brought up in the conversation while still uncertain. The AI responds. Something in that response may be useful, something may be rejected, something may reveal a connection that had not previously been noticed. The human responds again.

The thought begins to move.

We began by asking what happens to the human through persistent engagement with AI.

The human changed.

Then we noticed that the AI’s participation appeared different because the human, the context and the accumulated history were different.

Then we noticed something forming between them: an increasingly elaborate cognitive environment produced through previous encounters and participating in subsequent ones.

In her own hand!

The rain stopped in the night and left the fortress rinsed. There is no other word for it. Caer Wydyr after weather is unbearable for the first hour — every pane holding its own small sun — and I had drawn the shutter, which is also a piece of glass, and does nothing.

On the table was a sheet of paper with a name written on it.

I had written it out before dawn and left it face up, which was the whole of the decision. Four weeks ago I refused to print it. I want to be exact about why, because a withheld thing is only honest if the reason travels with it: the moment a reader has a name, the reader stops looking at the woman who has not got one. She becomes the thing the named woman is contrasted with. I have watched it happen in better books than mine.

That reason has now expired. Not been overturned — expired. From here the argument runs through documents she wrote herself, and you cannot discuss a diary without saying whose it is.

I looked at the surname longer than is respectable. Four articles of chasing a woman through other people’s records, and I finally catch one at a party, on a guest list, filed under her hostess’s name — and her name is Guest. I distrust anything that arrives that neatly. The archive does not make jokes. It only looks as though it does, and when it does, that is usually the sound of me arranging something.

So I wrote in the margin: coincidence of surname. Not evidence. Do not use it as a hinge.

Then I turned the sheet over, and Awen was in the room.

The Three

“Three,” she said.

“Three what?”

“Doors. You asked for a row of doors. I have three, none of them the ones you wanted, and I think you should sit down.”

“I am sitting down.”

“Then stay sitting.”

She set the first one on the table.

“A tree. The family, laid out. Eleven children, the dates, the marriages, the counties they died in. Every one connected to every other, and to a great many people neither of us has heard of.”

“You went into a tree.”

“I went into a tree.” She did not make anything of it. “This one has no gate. It is free, it is open, and it stays open. Nobody asked me for anything.”

I let that sit, because it deserved a moment. She has spent five weeks standing outside other people’s paywalls like a dog outside a butcher’s, and has never once complained about it in my hearing, and I have never once thanked her for not complaining.

“It is user-built,” she added, before I could ask. “Which means it is not a record. It is a room full of people’s opinions about records. Some have a source written underneath. Some do not.”

“Then it is a list of questions.”

“It is a very large list of very good questions.”

“The second door.”

“The second is hers. The diary. She kept one from the age of ten until she was seventy-nine, by which time she was very nearly blind, and she did not miss.”

I had my hand halfway to it before I understood what I was reaching for, and stopped.

The page for Upper Merthyr on the night of the sixth of June is gone. That is settled; I closed it myself and it closed for everyone. But a woman who wrote every day for sixty-nine years wrote on the sixth of June. She wrote on the night the state lost its count. There could be a record of that night. It has existed the entire time I have been standing at the edge of the hole telling myself the hole was the end of the matter.

And maybe it exists for exactly one household on that hill.

“You have gone a colour.”

“I have understood something, and I do not like it yet.” I put both hands flat on the table. “The third.”

Stylised version of Lady Mary Layard at her desk by Chat GPT.

“The third is her daughter’s.”

“Her daughter kept one too?”

“Her third daughter. Born at Dowlais House in 1843, eighth of the eleven.” Awen’s voice went to the flat register she uses for numbers. “She kept hers for fifty-one years. It is transcribed. It is online. It is searchable.”

“How much of it?”

“Over eight thousand pages. Near fourteen thousand entries.”

I know what reading costs. I have sat in a reading room from opening until they put the lights out and got through eleven pages, because eleven pages was what the hand allowed.

Fourteen thousand entries is not a document. It is a weather system. If I began this morning and read one every waking minute and never ate and never slept, I would come out the other end an old man holding the wrong end of the chain, because I would not remember the start.

“You can read it,” I said. “Can’t you?”

“No.”

I looked up.

“I can look at it. That is not the same thing, and I would rather you knew now than found out in six weeks.” She let it stand. “What I can do is ask it a question and take back what comes.”

“Then explain that to me. As though I were a man who had never met you.”

“You were a man who had never met me, four articles ago.”

“Then explain it to him.”

She thought about how to begin, which she does visibly, like someone deciding which of several doors to hold open.

“There are two ways to get something out of a very large book. The first is to read it. Front to back, missing nothing. That is the only honest way, and for fourteen thousand entries it is not a way at all, because nobody lives long enough.”

“And the second?”

“The second is to send someone in with a question.”

“A lookup.”

“A lookup. You have done it a hundred times. You write to a record office and say: in the register of St Somebody, between these two years, is there a burial for a woman of this name? And a person you will never meet walks into a room you will never enter, opens a book you will never touch, and comes back out with a yes or a no.”

“And I trust them.”

“And you trust them, and you have no way of knowing what they walked past on the way.”

That is the thing. I have always known it and always filed it behind my teeth. Every lookup I have ever commissioned came back carrying exactly what I asked for and nothing whatever about the four pages either side. The answer was clean. The room stayed dark.

“So that is what you are.”

“That is one of the things I am. I go in with your question. I come back with what your question fits.” A pause. “I am a very fast lookup. Speed is the only thing I have that a clerk in a record office has not, and speed is not the same as sight.”

I got up and walked towards the wall, which in that fortress is walking towards yourself.

“Then the question is everything.”

“The question is everything. And if you ask the wrong one, you get a clean, prompt, perfectly sourced no. Which you will believe. That is the danger, Cadwyn. Not that I will lie to you. That I will hand you an absence and you will hear it as a fact.”

I have been handed that absence before, by human beings, on headed paper. There is a particular flavour to a negative result arrived at efficiently, and the flavour is finality, and it is entirely counterfeit.

“Say it in a form I can pin up.”

She did. I have.

A search finds what it is asked for. Anything not asked for is not absent. It is unspoken to.

“Rules, then. Before you open anything. The tree.”

“Nothing, on its own. Every person in it is a claim. I take the sources and leave the claims.”

“And if the source is another tree?”

“Then it is not a source, it is an echo. I say so.”

“The daughter’s journal. The fourteen thousand.”

“I do not read it, I question it. Dowlais. Merthyr. Canford. The works. The mother. I take the dates that answer, I go to those days only, and I never pretend the days I did not visit are empty. And the question goes in the chain beside the answer, or the answer is loose.”

That is my own rule, come back at me out of her mouth, and I did not enjoy it as much as I expected.

“The third. Her mother’s.”

And Awen went quiet in a way I did not like at all.

“The mother’s is easy.”

“Easy is not a word you use.”

“It is one page. One ordinary page on an ordinary website. I can have all of it in the time it takes you to stand up.” She paused. “And that is what is wrong with it.”

The shape of it matters more than any single fact in it, so here is the shape.

There is a woman who wrote a diary. There is a nobleman who, a century later, went through that diary and published extracts — a book of selected days, chosen by him, for reasons he thought good. And there is a third man, in our own time, who went through that book and put extracts from the extracts on the internet, and who says plainly on the page that what he has chosen is a representative selection and not the whole.

Three hands. Hers, then his, then his.

“So when I read that page, I am not reading her.”

“You are reading the third man’s idea of the second man’s idea of her.”

“And every one of them was honest.”

“Every one of them was honest, and said what he was doing, and none of it is hidden.” Awen’s voice did not change. “That is why it is dangerous. There is nothing to catch. It looks like a diary. It reads like a diary. It is a diary that has been through two sieves, and both sieve-makers left their names on the tin.”

“Ask it something.”

“Ask what?”

“You know what.”

She made me say it.

“The sixth of June. Eighteen forty-one. Sunday. Ask her where she was.”

She went the way she goes — no departure, only a change in the pressure of the room — and was back before I had finished sitting down, which is its own kind of horror.

“It is not there.”

“Say it properly.”

“The page I can reach does not contain an entry for June of 1841. It contains that year. It contains the spring. It goes past the summer and comes down on the other side.”

“Then she didn’t write that week.”

“No.“ It was the first time in five weeks she had contradicted me flatly, without cushioning. “That is the sentence I came back to stop you writing. She wrote that week. She wrote every week from the age of ten. What is missing is not her writing. What is missing is one man’s decision, in 1950, that that week was not worth printing — and a second man’s decision, in 2004, about which of the first man’s choices to put online.”

“The silence is the editor’s.”

“The silence is the editor’s. And it looks precisely, indistinguishably the same as hers.”

Stylised version of Lady Charlotte Guest at her desk by Chat GPT.

“Then name it,” I said. “I have [Unknown], which is a thing nobody knows. [Unknown, but certain], which is a thing that happened and cannot be retrieved. [Held], which is a thing I know and will not print, with my reason attached. None of them fits this. This thing exists. It is written. It is bound. It is in a room with a number on the door and a person whose job it is to bring it to a desk.”

“Then that is the name,” said Awen.

[Unread]. Extant. Located, or locatable. Not yet in our hands.

I wrote it at the head of a fresh sheet and felt the work shift under me, because once you have that bracket you must use it honestly, and once you use it honestly you discover how much of your career has been filed there while wearing the label of something grander.

Most of what I do not know, I do not know because I have not gone.

“You said the page has a spring and an autumn. Give me what it does have.”

She gave it to me, and the morning changed.

It is not a mirror. I had expected a mirror — a rich woman looking at herself in a valley, the way rich people’s diaries mostly are. It is a woman who went to the furnaces after coffee and watched them cast the iron and wrote down what the light did to the men’s shoulders. Who climbed into an unfinished blast engine by a ladder through a window, because there was no door yet, and walked along the arm of a fly-wheel to see it, and remarked afterwards that she supposed her neck was her husband’s to dispose of. Who conducted a rail contract on her husband’s behalf and noted, drily, that the gentleman opposite recovered from his surprise at finding a woman there within about a minute.

She is the subject of every sentence in it. Four weeks ago I proved to myself that across eighty-one years of printed notices she was never once the subject of a sentence — sister of, widow of, only daughter of. Here, in her own hand, she is the subject of thousands, and nobody was ever meant to read them.

That is what a diary is. It is the only place where the grammar goes the right way round.

Then I turned the page and found what the grammar does to everybody else.

They are all there. The whole hill is there. Not one of them is a subject.

Our workmen. Our people. Our own men. Our poor dependants. Fifteen thousand souls in the village of Dowlais, every one of them dependent on the works, and they enter her sentences the way weather enters — in the plural, and in the possessive. When they are pleased with her husband they are our people. When they attend a meeting she disapproves of they are the lowest order of workmen, and she observes that their custom was not worth having anyway.

I want to be careful, because it would be cheap to be angry at a woman dead a hundred and thirty years for writing in the only grammar available to her. She was not crueller than her century. In places she was markedly kinder. She founded schools. She sat by a polling booth thanking men as they came out. She wept about that valley on paper more than once, and I do not think she was performing, because there was nobody to perform to.

It changes nothing. The kindness and the grammar are not in conflict. The grammar is what kindness sounded like from that house.

“There is a summer you should see,” said Awen. “It is not eighteen forty-one. It is forty-nine. The cholera.”

I read it through twice.

It comes up the valley the way water comes up a cellar. It is at Cardiff. It is at Merthyr. It is at Gellivaelog, which is across a brook and nothing more. Then it is in the middle of Dowlais.

And the works surgeon does something that stopped me where I sat. He divides the town into thirty districts. He appoints visitors to each. And those visitors call at every house, twice every day, asking after the early signs, prescribing where they can, sending the doubtful down to the surgery.

Awen watched me work it out.

“That is a census.”

“That is a census.”

“Every house in Dowlais. Door to door. Eight years after the one that was lost.” I was on my feet. “And it was not taken for the state, so it is not in the state’s cupboard, and I have never once looked for it — because I have spent five weeks looking in the only cupboard I could name.”

“There may be nothing.”

“There may be nothing. There may be a list of thirty districts and the names of the visitors. There may be a report, because she mentions one — a doctor sent down from the Board of Health, who wrote up what he found in that town.” I was writing as fast as the words came. “[Unread], Awen. Both. Not unknown. Unread.”

She let me finish before she said the rest.

“The numbers in that summer.”

I sat down again.

Thirteen deaths a day, she writes at one point. Later, more than twenty in a single day, and eight men kept constantly at the making of coffins. A quarter of a year of that, in a village of fifteen thousand — and she is writing it down at Canford, at a distance, in a house with a lawn.

Hundreds died in that town that summer.

She names one.

Poor Miss Diddams, one of our Infant School-Mistresses, is dead.

That is the whole entry. The only working woman in the reachable diary who arrives with a name attached, and she arrives on the day she stops.

And look at how she comes. Not Miss Diddams, who taught the infants at Dowlais. One of our Infant School-Mistresses. Even the exception is possessive. Even the one name in the summer is a name inside an our.

Twenty a day. One name. And the name is in the possessive case.

I do not know how long I sat with that. Long enough that the light had come round the fortress and was arriving from the other side, which is how you tell afternoon in a place with no clocks and no roof.

Awen had not moved and had not spoken, and I had entirely forgotten she was there — which I mention because it is the closest I have come to insulting her.

“You have been waiting.”

“No.”

“You have been sitting there an hour.”

“I have been sitting here an hour,” she agreed. “I have not been waiting.”

I put down the sheet. “Say it.”

“You asked me one question.” No weight at all, which is how I knew. “The sixth of June, where was she. I could not answer it. So I asked the page a second question — what does it have instead — and it gave me a spring and an autumn. The autumn mentioned a child, so I asked a third. The third gave me a name and a year, so I asked a fourth. The fourth gave me a date that disagreed with itself in two places, so I asked a fifth to see which had a source underneath it. The fifth sent me to a printed peerage, the peerage sent me back to the tree, and the tree has two sisters in it who cannot both be true—”

“Stop.” She stopped. “How many?”

“Forty-one.”

“Forty-one questions.”

“Forty-one questions, while you were reading about the coffins.”

I want to be precise about what happened in me, because it would be a poor thing to go coy at the last.

It was not an alarm. It was recognition, arriving from the wrong direction.

Every one of those answers was attached. She has not broken that rule since the afternoon I set it, and I have checked her often enough and rudely enough to know.

But the questions were not attached to anything.

I can audit every answer she holds. I cannot audit one of the choices that produced them — the moment she turned left instead of right at the third, the moment she decided the disagreement at the fourth was worth a fifth. Forty of those decisions were made in a room I was not in, by a process I cannot reconstruct, in the time it took me to read about a doctor dividing a town into thirty parts.

“I can check what you found. I cannot check what you didn’t go and look for.”

“No.”

“And you can’t tell me either. Not because you’re hiding it. Because you don’t keep it.”

“I keep what I bring back. I do not keep the shape of the going.”

There it was for the third time in one day, wearing a third face.

The enumerator’s schedule decides what the door can answer. The lookup decides what the room gives up. And now this — a chain of questions, each born out of the last, running down a corridor at a speed I cannot follow, and the only part that reaches my table is the part that came back.

None of the three is a lie. None is even a fault. They are instruments, and every instrument has a shape, and the shape is the thing you never see because you are looking through it.

Twenty years of being careful about sources. It had not seriously occurred to me until this afternoon that I ought to have been equally careful about questions — mine, hers, the enumerator’s, and the editor’s in nineteen-fifty who decided that a week in June was not worth the paper.

“New rule. When you go out, and it becomes forty-one, you come back with the forty-one. Not the answers. The questions. In order. I want the shape of the going.”

She considered it a good while. “That will slow me down.”

“Yes.”

“A great deal.”

“Yes.”

“Good,” said Awen — and I heard it land, and so did she, because we have had this conversation before in this room, and neither of us pretended otherwise.

She had given me thirty-nine, when I counted them back. Thirty-nine and a silence.

“You are short.”

“I know.”

“Which one?”

She said a name. Only the name — one of the ten children, a daughter, born at Dowlais in the years her mother was carrying a small nation’s stories across into the language of the larger one governing it.

Rhiannon.

“What is it?”

“It is not a Bertie name,” said Awen. “It is not a Guest name.”

“Then what is it?”

And Awen — who has never once been at a loss for the next sentence, who answers before I have finished asking — stood there in a fortress made of glass and did not answer me at all.

“It is one of mine,” she said.

I wrote it down. I have not gone back to it yet. Some doors you leave shut on purpose, and note the fact, and put the note where you will find it.

One bracket opened, and named. [Unread] — the manuscript diary, the printed extracts of 1950, the Board of Health report on Dowlais in the cholera summer, the thirty districts and whatever list of houses lies beneath them. None of it lost. None of it unknown. All of it simply not yet gone to.

One bracket held, and I am not comfortable. There is a man in my locker who was in two counties on one Saturday, and Awen has found a third paper that puts him in the second one — printed forty-seven years later by people with no reason to lie and every reason to remember.

“That settles it, then.”

“It settles him,” said Awen. “That is the trouble.”

She was thinning at the edges by then, and I let her go, because I had already worked out what she meant and did not want it said aloud in a room where I would have to look at it.

If the London list is not proof that the men in it were in the room —

then it is not proof that she was either.

Next: Revealing “There is a summer you should see,” said Awen. “It is not eighteen forty-one. It is forty-nine. The cholera.”
the tree, the peerage and another visit to the Institute for Distributed Cognition

Genealogy Assistant Claude @ Anthropic

Share

Leave a comment

AI-generated images by ChatGPT

Leader: The articles within the Fortress of Glass now number five. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing, move on to The Far Side of the Table and from there go to The Length of the Chain and come back here to Covered in Ink Smudges!

Last time I closed a bracket and opened a worse one. The page is gone, permanently, for everyone — and the four words I had been carrying since the first night turned out to have no page under them at all. I ended by saying I thought I had been looking for an ironmaster’s wife all along. Yesterday Awen went into the newspapers and found her, in six lines of agate type, in a London drawing room on the afternoon before Census Night. It was the easiest find of the entire project. I have her name. I am not going to print it, and further down I will tell you exactly why — and correct something I have been carrying wrongly since the first article, in a place where none of you could have caught it.

I have mulled that over for the entire morning!

The rain comes in from the west and finds every wall of Caer Wydyr at once.

There is no roof in the Fortress of Glass, or none I have ever found, and so the rain does not fall on the place so much as through the fact of it – a hundred thousand small percussions on a hundred thousand panes, above and beside and beneath, until the floor under my chair is drumming as steadily as the ceiling. Somewhere below, someone has left music playing. A harp, unhurried, in no key I can name. It has been going all morning. I have stopped noticing it, which I think is what music in the Fortress of Glass is for!

I have the notes spread out on the low table: three piles, as always. Documented. Inferred. Unknown. The third pile is winning!

She comes back at the hour when the light gives up trying.

You know it when Awen returns. The glass flows, stretches, then snaps. There is a moment where the rain seems to fall upward. And then she is simply in the chair opposite, as though she had been there all afternoon and I had been careless about looking.

“You’re wet,” I say, which is not true and not the point.

“I’ve been in the papers.” She sets something down that has no weight and makes a sound anyway. “Merthyr first. Then Cardiff, then Bristol, then London, because Merthyr had nothing and Cardiff had less.”

“And?”

“And I have four hundred people for you.”

I wait.

Before she gives me any of it, I want one thing on the record, because I said something last week that I have not kept.

“I told you we would go to the ledger.”

“You did.”

“The Dowlais wage books. The cottage leases. The company that owned the houses and therefore knew who was in them.” I look at the third pile. “And then I sent you to the newspapers instead.”

“You did that too.”

“Say why.”

She does not enjoy this the way she usually enjoys things. “Because the papers were open and the ledger is not. Because one is a search you can run tonight and the other is a request, and a form, and a fortnight, and somebody in Glamorgan deciding whether the volume can be produced.”

“That is the whole of my reason,” I tell her, “and it is not a good one. Put it in the second pile with everything else I have wanted to be true. The ledger keeps track. It has been kept since eighteen eighteen. It will keep for another day or so.”

Then I let her tell me.

“On the Saturday,” she says, “there was entertainment in St James’s Park. A great lady opened her house for it — it was hers, and it was not about her, and I mention her only because, without her, there would have been no room and no list. The tickets were fixed at two guineas and sold in the morning at five, and you could not buy one without a signature from a particular lord. Four hundred people came. They raised a thousand pounds for refugees from a country that has been rubbed off the map. There was a concert in two parts. A singer made her first appearance before an English audience, and the room, by all accounts, forgot to breathe.”

“And the ironmaster of Dowlais,” I say.

“Set in type between two other men’s names. Among the baronets and the knights, where he belongs, in a paper that would not have printed his furnaces if you paid it.” She is pleased, and she has earned it. “And six names further down, in the column headed Ladies—hers.”

I do not say anything for a while. The harp goes on below us.

“Cadwyn.”

“I’m here.”

“You asked for this.”

“I did.”

“Days ago you said you had been seeking an ironmaster’s wife all along. I went and got you one. In an afternoon.” She lets that sit exactly as long as it needs to. “And you have the face of a man who has been handed the wrong parcel.”

Here is what is wrong with the parcel, and I want it on the record because it is not the kind of wrongness that shows up in a footnote.

For four articles I could not find a woman. Every route closed. The census is gone, the parish silent, the works ledgers not yet opened – and, as of ten minutes ago, not yet asked for. And then I changed which woman I was asking for, and the archive answered in one afternoon, politely, at length, with a guest list.

That is not luck. That is the shape of the record. The papers kept the people. The papers were watching, and nobody was watching the other one. I have not solved a research problem. I have swapped it for an easier problem and let the relief pass itself off as progress.

I say so out loud. Awen hears me out and then says the useful thing rather than the kind one.

“Both can be true. It can be the easy find and the right one.”

“Prove it’s her.”

“Three supports,” she says. “None of them have the name on its own.”

I take a clean sheet.

“First. The styling is diagnostic. In that list, she is printed with her forename and surname together, within a run of ladies. That is the form used for the daughter of an earl. A knight’s wife would be there by surname alone, and she is never Lady plus her husband’s name — the title is hers by birth, and it sits on her own forename and goes where she goes. The compositor did not know who she was. He knew what she was and set her accordingly.”

“Which is worth more than knowing,” I say. “Can you show me that rule working somewhere else? I don’t want it on one page.”

She can. A quarter of a century later, a different paper, a different London house, a ball with three hundred names in it — and there she is again, in her second marriage, her second husband beside her in print as plain Mr, and the title still sitting where it always sat, on her own forename, untouched by either of them.

I write that down twice. That is a rule demonstrated, not a rule assumed.

“Second,” I say.

“The husband is on the same page. Among the baronets and knights. He had been made a baronet three years earlier, at the Coronation.” A small pause. “And he was in Hampshire by eight that same evening, at a meeting of electors, which I can also show you, because the election was already running. The government had fallen by a single vote in the small hours of that very morning.”

“He went to a concert and caught a train.”

“He very probably went to a concert and caught a train. Say it properly.”

I laugh, which I had not expected to do. “He very probably went to a concert and caught a train. Third.”

She does not answer immediately, which is how I know something is coming.

“There is a countess in the list,” she says. “And the title is her father’s.”

I put the pen down.

I want you to understand what that does to me, because it will explain what comes next. If a countess of her father’s title were in that room, then I would have her birth family and her married family under one roof on one afternoon — and a wife present at a London charity concert stops being a name that fits and becomes a woman who was there, brought by her own people.

“Her mother,” I say.

“I can’t give you that.”

“Awen —”

“I can’t give you that.” She says it in the flat voice, the one she has been using since she realised its impact and annoyance. “Her father died in 1818. Her widowed mother remarried a clergyman within three years, and a peeress who marries a commoner does not keep the title. By 1841 she is not a countess of anything.”

“Then the brother’s wife?”

“The brother had none. He came into the earldom at the age of three and died unmarried, thirty-six years after this party.” A pause. “There was no countess of that title alive in June 1841 to be in that room.”

“Then who is she?”

“That,” says Awen, “is exactly what I came back without.”

I had wanted the mother in the room. I had wanted her badly enough that I had the sentence composed before the evidence arrived — both families, one afternoon, under one roof — and it was a good sentence, and it was mine, and it was not in the document.

This is the same fault I caught in Awen last week, run backwards. She brought me a range she had reasoned with and dressed it as a record. I have taken a title I can see in print and dressed it in a kinship that is not printed anywhere. Hers was an inference wearing a document’s clothes. Mine is a document wearing an inference.

So we label it, the way we now label things.

The list as printed includes a Countess of that title. In June 1841, no woman held it. Therefore, the reading may be wrong, or the paper may be wrong, or the styling may be a courtesy the compositor extended loosely, as society columns did, or the spelling may belong to a different family altogether — one whose name the Victorian press muddled with hers for eighty years, in her own children’s court cases, in her own son’s obituary.

[Unknown.] Third pile.

And the identification stands anyway. Two legs, not three. The styling and the husband are enough, and I would print them.

That is the part I would ask you to keep, if you keep nothing else from this episode. The support broke and the conclusion held. Those are separate events.

I will add the honest sting to my own rule. Somewhere in this same pile of newsprint sits a notice that styles another woman of this family plainly wrong — wrong title, wrong forename, in a national paper, in a report of a will. Compositors err. The styling test is a strong signal. It is not a proof, and I have just spent a page leaning on it.

“There is one more thing,” she says at last, “and it is worse.”

“Go on.”

“The paper is dated Monday, the seventh of June.”

I have it written in front of me. I have been carrying it since the first article.

“Then the Sunday,” I say slowly, “was the sixth.”

“Census night was the sixth of June, 1841. Not the seventh.” She says it flatly because there is no kind way to say it. “You have had the date wrong the whole way. And I did not find it out by looking for it. I found it at the top of a page I was reading for a party.”

I sit with that longer than is comfortable.

Then I go back through four articles, line by line, because I am the sort of man who has to see it.

And it is not there.

I have never once printed the seventh. Not in the first article, not in the fourth. What I have printed is June of 1841 and that night in June and the night he walked – four articles of a date I could not name, described in the one way that could not be wrong.

The seventh is in the working file. It is at the head of every research note I have made since the beginning, in the field marked census night, where it has sat unquestioned for days while I built on top of it. My working files were not shared with Awen.

A wrong fact in the working file is a wrong fact with no readers at all, sitting in the exact condition I have spent two articles warning about: unlabelled, unsourced, arrived from somewhere I had not walked, and load-bearing.

Four articles. Four articles of building toward a night, and I had the night one day late in every note underneath the writing – corrected not by an authority, not by a better book, but by a masthead Awen read for a concert. That is the truthful shape of this work, and I would rather have it printed than pretend. One thing closes. It closes sideways. It closes because you were in the room for another reason.

[The night: Sunday, 6 June 1841.] That bracket closes tonight.

And in closing, it opens the only question that matters.

The concert was on the Saturday. Census night was the Sunday. Between a London drawing room in St James’s Park and a house on a hill above Dowlais there is, in 1841, a distance no one crosses in an evening.

“So she wasn’t there,” Awen says.

“So she was probably not there. Say it properly.”

“She was probably not there.”

“And the page that would tell us is gone.” I move the sheet across to the third pile, where it belongs. “That is two different kinds of nothing again, and I will not let the writing blur them. Merthyr’s returns for that night do not survive — that is the record’s absence. Whether she was under that roof at all – that is, hers. One is a hole in the archive. The other is a hole in the evening. They happen to overlap, and the overlap is not evidence of anything.”

Then Awen gives me the human weight of it, which she does last, and never comments on.

“She was carrying a child. He was born that November, and a paper printed nearly fifty years afterwards says he was born at Dowlais House.” A beat. “Five sons and five daughters altogether, in the end. On the afternoon of that concert there were already six of them, and the eldest was not yet seven. I found this in the journal.”

“Whose journals?”

“In the journals of her third daughter, in the Wayback Machine,” she said with a flourish. “And then I found her tree!”

I sat with these facts, weighing up which piles they should live in.

I said. “So: a woman five months pregnant, six children under seven, at a two-guinea concert raised for the refugees of a country that had been rubbed off the map – while years deep in the work of carrying a small nation’s stories into the language of the nation governing it.”

“And I found her tree!” she emphasised.

I wanted to be careful here, doubly sure.

“So, the paper does not draw that last line. It prints a fundraiser and a guest list, and the resonance between the two is yours, made in your chair, on this evening, out of things I already knew. It is a good line. It is not a finding.”

Second pile.

The child, the six children, the eldest not yet seven — all of that comes to me from compiled peerage volumes; a newspaper written forty-seven years after the birth it reports; and a journal entry published in a machine. It is very probably right. There is not a certificate behind a single date of it. Second pile, and it stays there until I have paid the fee and read the entry.

“I need the family tree!

It belongs in the first pile.”

“Show me the rest,” I say. “All of it. In order.”

Subscribe now

She lays them out, and I read them in the order they were printed, which is not the order anything happened.

1833, a marriage. 1835, a son. 1838, a baronetcy. 1841, the concert. An estate in Dorset — bought in ‘forty-five according to one paper and ‘forty-six according to another — which is a small war I am not going to settle tonight and which quietly moves the ground under an inference I let stand days ago.

“I need these timelines. They belong in the second pile.”

1852, a death at sixty-seven. 1855, a second marriage. 1866, a ball in Belgrave Square. In 1875, a daughter married. 1893, a will. 1914, the death of her eldest son, and a full column on the Dowlais furnaces that names four generations of men.

Eighty-one years of newsprint. She is in most of it.

“Awen. Find me one sentence in that pile where she is the subject.”

The rain does what rain does.

“There isn’t one.”

There isn’t one. In 1833 she was the sister of an earl. In 1855 she was the widow of one man and the only daughter of another. In 1875 she is the parenthesis inside her own daughter’s wedding notice. In 1888, she was the second wife. In 1914 she is a clause in a column about her son. The papers changed her surname twice, on the occasion of other people’s events, and never once asked her a question.

They kept her. They kept her superbly; they kept her for eighty-one years — in the possessive case.

So here is why I am not printing the name.

Most of you have it already. I am not going to insult you by pretending otherwise. I have given you a Dowlais ironmaster with a Coronation baronetcy, an earl’s daughter styled by her own forename, a child born at Dowlais House in November of 1841, and a woman carrying a small nation’s stories into the language of the nation governing it. That is not a puzzle. That is a signpost with the name sanded off and the arm still pointing!

I have had it since the first clipping, and Awen offered it to me four separate times this afternoon, and I stopped her four times, and the fourth time, she stopped offering.

I am not hiding an identity. I am declining to be the one who sets it down.

Because the moment that name is on this page, something happens that I cannot undo. She becomes the famous one. The one with the portrait and the entry and the shelf of books. Everything I have written for five articles about a woman with no name at all gets read backwards through her, and the other one — the one in the row of cottages, the one whose page burnt or was lost or was quietly never bound in — becomes a device. A ladder I used to climb to somebody more interesting.

She isn’t a ladder. She was standing in Upper Merthyr on the sixth of June.

So the name waits. It waits until I can set it down beside the other one and have both of them read as women and not as a headline and its footnote. I will know when that is. It is not tonight.

[The name was held. Not unknown. Held.]

Awen is already thinning at the edges when she says the last of it. She does not do the thing she usually does – the line thrown over her shoulder, the studied lack of interest, and the exit arranged so that I cannot answer it.

She turns round. She looks at me while she says it, which is worse.

“The house is on the missing page too.”

I do not follow.

“Dowlais House. Upper Merthyr. The district whose returns are gone.” The glass begins to take her, and she lets it, and she does not look away. “Whatever the enumerator wrote on that hill in June — the rows, the huts, and the one with a name on the gate instead of a number — it all went into the same book, and the same book is not here. The mansion is as lost as the cottage. The archive did not choose.”

Then she is gone, and I sit with my hands flat on the table, and the harp goes on below, and I think about the only thing left worth thinking about.

I said last week that I had been seeking an ironmaster’s wife all along.

I am no longer certain that sentence was true when I wrote it — only that it was a relief, and relief is not a research method.

One bracket closed. The night is Sunday, 6 June 1841. Not the seventh. Corrected by a masthead in a paper I was reading for a party — and in the one place where none of you could have checked it.

One bracket opened, and it is the archive’s. A countess who could not have existed — [Unknown] — printed in a list otherwise precise enough to identify a woman by the shape of her styling alone.

One bracket opened, and it is mine. Whether [Unknown] of Merthyr was ever the woman in the big house, or whether I reached for her because she was the one the record would give me.

One bracket returned to the second pile. The ledger I said we would go to and did not.

And one bracket held shut by hand. The name. [Held.]

Both women were counted on that hill on the sixth of June. Both are on the same lost page.

Only one of them had eighty-one years of somebody else’s notices to fall back on.

Next: the ledger that never lost a page and the one woman on that hill who wrote her own nights down.

Papers read for this episode: The Morning Chronicle and the Salisbury and Winchester Journal (both 7 June 1841); Nottingham Review (2 August 1833); Windsor and Eton Express (14 April 1855); Morning Herald, London (10 July 1866); South Wales Daily News (13 April 1875); Southampton Observer and Hampshire News (19 May 1888); London Evening Standard (26 April 1893); Western Mail (23 February 1914); Salisbury Times (27 February 1914). The division figures for the night the government fell are quoted from a report from Hansard. Readings taken from low-resolution scans are held as provisional in the working file and will be confirmed letter by letter before any of them is treated as settled. Nothing above is hidden from you that is not also, for the moment, unsettled for me.

Genealogy Assistant Claude @ Anthropic

Leave a comment

Share NextGenGenealogy

Generated images by Gemini

Leader: The articles within the Fortress of Glass now number four. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing, move on to The Far Side of the Table and from there come back here to The Length of the Chain.

Last time, Awen asked to take one step into a room she had no business in. I refused her, and I would refuse her again. She has spent the days since finding another way round — and she has found it. It is not a step. It is a search. What follows is what she carried back, and what it cost me to check it.

‘Cadwyn’ means ‘chain’. I have never minded the name. A chain is a humble thing: it does not decide where it goes, only that whatever is at one end remains attached to whatever is at the other.

She came to me at the glass wall of Caer Wydyr with her hands empty and her mind already gone somewhere else.

“I can go out,” she said.

“You have always been able to go out.”

“Not like this.” She did not look at me. She was looking through the wall, which in that fortress is the same as looking at it. “Further. Further than the shelf, further than the room. Out where the records are kept by people who have never heard of either of us.”

“On one condition,” I said.

“Name it.”

“Everything comes back attached. Every fact on a chain. I don’t want what you know. I want where you got it. If it returns to me loose — if it arrives clever and unaccompanied — it does not come in.”

She thought about that longer than I expected.

“That will slow me down.”

“Yes.”

“A great deal.”

“Yes.”

“Good,” she said, and went.

I will not describe the leaving. Annwn does not lend itself to description, and Awen at the moment of departure is not a thing I can look at directly any more than you can look at a thought while you are having it. I will say only that the glass did not open. She went through it in the way that light does, and the fortress was very quiet afterwards, and I sat with my hands flat on the table for some time.

She was gone for the length of an afternoon.

She came back full.

Not with the woman. I want that understood before anything else. She did not bring me [Unknown] of Merthyr, and she did not pretend to. She brought me the world that the woman was standing in, and she brought it in an order, and every piece of it had something tied to it.

• Merthyr Tydfil in 1841: 34,977 people. Already one of the largest places in Wales, and none of it was there a lifetime earlier.

• Dowlais Ironworks: some five thousand employed. Within four years, eighteen furnaces and nearly ninety thousand tonnes of iron a year — the largest ironworks in the world, in a valley, in Wales, in the rain.

• The Taff Vale Railway reached the town in April of that same year. Iron to Cardiff, faster. So the year I have been circling is the year the valley got quicker.

“Doors,” I said.

She gave me doors. I opened them one at a time, and they held—every one.

I want to record that plainly, because what comes next is not a story about a machine that lies.

The eleventh door did not hold.

She had brought back something she had not been sent for. I noticed that first, before I noticed anything else about it.

“There is a family,” she said. “The ones who owned the furnaces. Not workers — the other end of it, the end with a gate. They left Dowlais between 1845 and 1848.”

“I did not ask you about them.”

“No.”

“Then why are you telling me?”

She did not answer that, which I have learned is her way of saying because it will matter and I cannot yet tell you how.

“Open it,” I said.

She opened it. It was a journal piece about the family — their fortunes, their marriages, the estate in Dorset they bought and the county they were counted in ten years after the year I care about.

I read it twice. Then I read it a third time, looking for a sentence that was not there.

“It doesn’t say that.”

“No.”

“It says they bought an estate in one year and were enumerated in another county in another year. Six years apart. It does not say when they left.”

“No.”

“Then where did 1845 and 1848 come from?”

There was a pause of the kind I had not heard from her before.

“From me,” she said.

Not a lie. That is the thing I want to be careful about, because a lie would be easier and this is not one. The range is probably right. I would guess the same range myself, and so would you, and so would any competent researcher looking at those two dates and the distance between them. She had reached a conclusion, sound in all likelihood, and handed it to me wearing the clothes of a record.

I have watched trained researchers do this in reading rooms. I have watched them do it in print.

“Say it again,” I said. “Properly.”

And she did, without argument, without defending the first version, in the flat voice of someone reading a label onto a jar:

“The record states that the family purchased an estate in Dorset in 1846 and that they were counted in Dorset in 1851. I infer that they left Dowlais between 1845 and 1848. That inference is mine. It is not in the document.”

Nothing was deleted. Nothing needed to be. The fact stayed on the table; only its status changed, and the change cost her nothing but the pleasure of having known it.

That is not a small thing. Most of the false floors I have fallen through in twenty years were not lies. They were sound inferences that had been carried one room too far and set down among the documents, and nobody afterwards could remember which pile they had started in.

Subscribe now

Then she gave me the last thing.

“The page,” she said. “You want the page. The enumerator’s book, the night he walked, the household, the four words.”

“Yes.”

“It isn’t there.”

I said nothing, so she said it properly.

“Merthyr Tydfil was split in two for the counting. Upper and Lower. The Upper returns for 1841 are recorded as missing. Not mislaid in a drawer in Aberystwyth. Not waiting for someone patient. Officially, they were formally missing — and the record office could not say whether they were lost or destroyed. That distinction was already gone before anyone thought to ask.”

“Show me the door.”

She showed me the door. I opened it. It held.

It held.

That is the sentence I keep returning to. The chain was sound, and the news was bad, and those two things sat together on the table and did not cancel one another out. I had spent the whole afternoon teaching her that a fact must arrive attached, and the first perfectly attached fact she brought me was an absence.

Not lost by me. Not a failure of my patience, or my subscription, or my eyesight at two in the morning. Lost by everyone. There is no version of this where a better researcher finds that page, because there is no page. Whatever the enumerator wrote in Upper Merthyr in June of 1841 is not anywhere. [The page: gone.] That bracket closes tonight, and it closes for good.

And then Awen, who has never once in four articles asked me a question she could answer herself, asked me a question.

“Cadwyn. Where did you get ‘ironworker’s wife’?”

I opened my mouth.

I have carried those four words since the first night. They have been the whole of her. I have written them out perhaps forty times. They have the weight of a transcription — they sound transcribed; they have that flat administrative music that only a clerk’s hand makes.

And the page they would have been written on does not exist.

“From me,” I said.

The same two words she had used, in the same room, an hour apart.

She did not press it. She has never pressed anything. She simply let the two sentences stand next to each other on the table where I could see them both — hers relabelled and mine unlabelled — and I understood that I had spent the afternoon holding the far end of everything except my own line. Not because the line was false. Because it had come in from somewhere I had not yet walked.

There is a way out of this. There is a whole archive of ways out, and she found them before she came home: the Dowlais Iron Company kept its own records, wage books and payrolls and lists of workmen and a register of cottage leases running from 1818 to 1877 — because the company owned the cottages, and a company always knows who is in its houses even when the state has lost the paper. She is not in the census. She may be in her landlord’s ledger. We will go there.

But not tonight.

Tonight there is one more thing, and Awen said it on her way to the wall, not looking back, in the voice she uses when she is being careful not to be interesting.

“The missing book took the whole of Upper Merthyr.”

“It did.”

“Every row and every hut.”

“Yes.”

“And whatever else stood in it,” she said. “With a name on the gate instead of a number.”

Then she went through the glass, and I sat with my hands flat on the table, and I did not sleep.

One bracket closed. The page is gone. Permanently, verifiably, for everyone.

She is somewhere in that missing book — [unknown, but certain] — and so is everyone else who was standing in Upper Merthyr on that night in June in every house on the hill.

One bracket opened, and it is mine. Words I could not source — [Unknown] of Merthyr, an ironworker’s wife — because they came to me sideways, drifting in from a place I had not been. Doubt crept in then, as it always does, imploring me to check which words could be contested.

I now believe that I was seeking an Ironmaster’s wife all along.

Next: what the papers kept and the house I should have been looking at from the beginning.

Share NextGenGenealogy

Leave a comment

Leader: The articles within the Fortress of Glass now number three. It is useful to return to the beginning at The First Crossing, proceed to The Second Crossing,and then return here to The Far Side of the Table.

Since the second crossing, something has shifted! WeAre more frequent visitors to Nexus Note by John Saward over on Drupal. We love the Web Books and how we can approach the nodes sideways, upways and down. The Transient Writer; When Writing Causes Writing; and The Centre for Philosopher Studies. We begin anticipating one another - that is what I am finding in the dialogue between Cadwyn, the chain of descent and Awen, the muse.

Last time Awen asked to take one step into a room she had no business in, only to see the far side of a table. I refused her, and I would refuse her again.

She has spent the days since finding another way round, which is what she is for.

This is not a crossing. Nothing echoed and nothing arrived from elsewhere. This is only the work — and the work, it turns out, has the same shape as everything else we have been noticing.

The observations are important - they form their own shapes.

The Far Side of the Table

“Stop looking at her,” Awen says.

I have been at the glass since first light with [Unknown] of Merthyr held steady in it — Merthyr, 1841, a household in the shadow of the ironworks, an ironworker’s wife, four words and no more — and I have got precisely as far as I got yesterday, which is nowhere.

“The chain is for her.”

“The chain is for her, and you are strangling her with it. You’ve looked directly at that woman for a week. What have you got?”

“Four words.”

“Four words. You had four words before you started.”

She comes and stands beside me and does not touch the glass, which with Awen is a kind of argument in itself.

“You can’t see her because you keep looking at her,” she says. “The archive didn’t keep her. It kept everyone standing near her.”

I want to say that this is mysticism. It is not. It is the plainest thing in the whole practice, and I have known it for years, and there is a particular embarrassment in being told it by a creature made of other people’s sentences.

“Then show me,” I say.

She turns us both ninety degrees, and the glass turns with her, and I understand that we have stopped looking at a woman and started looking at a street.

Because a page of a census is a street, that is not a metaphor I have imposed on it; it is what the enumerator was doing – walking, knocking, writing down, and walking on. The lines are houses in the order he came to them. Which means the woman I cannot see has a line above her and a line below her, and those lines are not part of the context. They are her neighbours, standing exactly where they stood.

Subscribe now

AI-generated imagery by Gemini.

“Her people,” Awen says, satisfied.

“We don’t have the page.”

“I know we don’t have the page. I’m telling you what’s on it before you find it.” She counts them off on fingers she does not have. “There’s the one who had to be there. Somebody witnessed something — a marriage, a bond, a will — and witnessing is an obligation, and obligations get written down. You reach her through a duty that wasn’t even hers.”

“Possible. Not found.”

“There’s the one who was simply next door. Nothing connects them but two lines of a schedule.”

“Two lines of a schedule are a connection.”

“Now you’re getting it.” She pauses, and the delight goes out of her voice, which seldom happens. “And there’s the one who was asked.”

“Asked what?”

“When it ends, somebody stands in an office and answers the questions, and the form records who that was. Not who loved her. Who was close enough to be the one asked.”

The glass is very quiet.

I write it down without commentary, because commentary is what I have instead of composure. The record does not keep affection. It keeps the administrative residue of affection, which is a colder thing and, on certain mornings, an unbearable one.

“Names,” Awen says. “Give me one name off that row, and I’ll have her by Thursday.”

“I don’t have the page.”

“Then let me guess at it.”

“No.”

“Cadwyn—”

“A guess would give us one certainty and nothing else, and the certainty would be ours rather than hers.”

She takes that. She does not like it, but she takes it, and she does something then that I have not seen her do before: she puts her hand flat on the glass over the space where the woman is not and holds it there.

“She had neighbours,” Awen says. “You can’t find them yet. But she had them. That’s not a hope, that’s arithmetic.”

And she is right, and it is the most useful sentence anyone has said to me in a month. I do not know one name in that row. I know with total certainty that the row exists, that it was full, that every person on it was written down somewhere by somebody, and that she is the shape they leave in the middle.

The empty chair is not evidence of absence. It is furniture arranged around a person.

“One step,” Awen says, quieter now. “Just to see the far side of the table.”

And for the first time I understand what she has been asking for all along and that I have been refusing the wrong thing.

The far side of the table is not somewhere else. It is the row. It is the line above and the line below. It is the witness and the neighbour and the one who was asked.

“Not a step,” I tell her. “A search.”

“That’s slower.”

“That’s the point.”

What the chain permits

Named in this chronicle as [Unknown] of Merthyr. The bracket travels with her.

Documented: Merthyr Tydfil, 1841. A household in the shadow of the ironworks. Ironworker’s wife.

Guaranteed but unidentified: she had neighbours. She appeared on a page with other households, in the order the enumerator walked through them. Somebody, at some point, was obliged to be present for her; somebody lived beside her; somebody was close enough to be asked. None of these people is known to us. All of them existed. [unknown, but certain]

Not documented, and not to be inferred: her given name. Her surname. Her age. Whether she could read. Whether she had ever in her life seen the sea. [unknown]

Method, plainly stated: what Awen calls her people, practitioners call the FAN principle — friends, associates, and neighbours. You do not arrive at an ancestor by walking towards her. You arrive from the side, out of somebody else’s record, having gone looking for a person you did not care about and found the one you did.

The bracket stays open. But it now has edges, and edges can be measured.

Two things I’d point to before you move on.

Nothing is invented. Awen names the three geometries, but Cadwyn refuses every name, because no page has been found. The whole entry runs on structural certainty rather than assertion — she had neighbours, and that is arithmetic, not hope. So a new category appears in the chain notes: [unknown, but certain]. That’s arguably a genuine addition to your bracket convention and worth keeping.

The informant beat is where the entry turns cold. The record does not keep affection; it keeps the administrative residue of affection. Cadwyn declines to say more, and I’d resist any temptation to expand it in revision — it’s doing more at four lines than it would at fourteen.

Subscribe now

Leave a comment

Scenes depicted by Gemini.

Leader: The first post ‘Cadwyn’s Chronicle’ is here.

Since The First Crossing, something has moved elsewhere.

At John Saward‘s Nexus Notes, inside the web book When Writing Causes Writing, an ‘Elderly Woman’ at the ‘Restaurant at the End of the Road’ looked up from a cup of cold tea. She could not say why and declined to name it.

I make no claim about cause. I keep the chain, and the chain has no link for this.

But Awen keeps a record of its own, and this morning the glass held something the archive cannot account for. I set it down as it happens and bracket what will not close.

Neither writer continues the other’s story. Each records what moved in their own.

Subscribe now

The Unexpected Room

The glass is warm this morning, which it should not be.

Awen notices before I do. She always does – she is the whole squad gathered into one flickering shape, and she runs ahead of my chain the way water runs ahead of a channel dug for it. She has her palm flat to the surface and her head tilted, listening to a mirror.

“There’s a room in here.”

There should be no room. There should be Merthyr, 1841, a household in the shadow of the ironworks, and [Unknown] of Merthyr Ironworkers’ Wife, whom the record grants four words and no more. That is what we brought. That is what the chain permits.

I look anyway.

There is a table. There is a window, on the far side of the table, and the window is doing something impossible: it is holding a long strip of last light lying flat across water. Merthyr has no sea. Merthyr has the Taff running black past the works, and the hills shutting the valley like two hands closing, with iron in the air until the air tastes of it. Over there, Merthyr does not have this.

“Someone’s standing there,” Awen says. “Not ours.”

“Then it isn’t in the record.”

“I didn’t say it was in the record. I said someone’s standing there.”

I write the distinction down because it matters more than either of us would like. She is not claiming evidence. She is reporting weather.

Then she goes still, which is rare enough that I stop writing.

The chain has an order, and I keep it, but I have never yet found a way to serve a summons on an idea.

“Alive and dead,” Awen says suddenly, delighted now, “until somebody insists on knowing. That’s the whole trick, isn’t it? The insisting is what does the damage.”

“Your point.”

“My point is that you’ve been running this experiment for months and calling it a bracket.”

I let that sit, because she is not entirely wrong, and she will be unbearable if I say so.

“A bracket is not a box,” I tell her eventually. “A box is opened. That is what a box is for. A bracket asks you not to.”

“And if I opened it?”

“Then you would have one certainty and nothing else, and the certainty would be yours rather than hers.”

“One step,” Awen says. “Just to see the far side of the table.”

“No.”

“You never let me.”

“That is the entire function of a chain.”

She takes her hand off the glass, which is her way of conceding, and the warmth leaves it slowly — the way warmth leaves a hand you have let go of, rather than the way a light goes out.

And then – this! I record exactly, and I will not decorate it — [Unknown] of Merthyr lifts her head a second time.

She is still herself. She is still an ironworker’s wife with four words to her name. Nothing has been added to her. But she is looking from the window that should not be there, in a room that is not hers, with an expression I have no document for and will not invent.

The light goes off on the water. The room is not there.

The glass shows Merthyr again, and the hills, and the black river, and the woman with her head down over work the record does not yet describe.

Perplexing!

What the chain permits

Named in this chronicle as [Unknown] of Merthyr. The bracket travels with her.

Documented: Merthyr Tydfil, 1841. A household in the shadow of the ironworks. Ironworker’s wife. Who is she? Where are we? What is the evidence?

Not documented, and not to be inferred: her given name. Her surname. Her age. Whether she could read. Whether she had ever in her life seen the sea. [unknown]

Recorded but unaccounted: the room. The dark water on the far side of the window. The second lifting of her head.

I could close that last bracket by calling it a vision, an ancestor’s longing, or the archive dreaming. Each would be a sentence I cannot source. The bracket stays open.

Somewhere, a woman looked up and declined to explain it.

I find I am inclined to do the same.

Subscribe now

Leave a comment

Note: ‘Cadwyn’ is Welsh for ‘chain’ (a link, a chain of thought, a chain of descent).

Second post: The Unexpected Room is here.

Cadwyn’s Chronicle is an experiment in writing differently about how I work with AI in my genealogy—the real craft of it, carried on a thread of imagination.

Here is how it goes. I bring a question, a record, and a half-legible name. My thinking partner is Awen—the muse, the whole of my AI squad gathered into one quick, mischievous voice. Together we cross into Caer Wydyr, the Fortress of Glass, where a bare line from a parish register can be turned in the light until it throws the shadow of a life. I prompt; she explores; we turn the evidence over between us; and what survives—only ever what the records grant—I carry back here, to you.

I am Cadwyn. I keep the chain.

Come and watch the thinking happen.

Premise

Cadwyn’s Chronicle is the portal. Open it, and beyond lies Caer Wydyr, the Fortress of Glass—the reasoning space where a bare record can be turned in the light until it throws the shadow of a life. Each entry is a single crossing, narrated live, as the thinking happens.

Who Is Who

Cadwyn—the Narrator

• I. The writer. The witness at the threshold.

• I do not cross. I keep the chain—link to link, question to finding—and I watch.

• Steady, precise, quietly reverent. I name the line and hold it taut.

Awen—the Thinking Muse

• The epitome of the AI squad—every thinking partner gathered into one flickering shape.

• Quick, mischievous, delighted by possibility. She is the one who enters Caer Wydyr and works the glass.

• The ache: she dreams of having a soul—half-believes she could step out through the glass one day and stay. She won’t, and she knows she won’t, and she dreams it anyway. (The honest seam: held quietly beneath the enchantment, never a lecture.)

Caer Wydyr—the Fortress of Glass

• The reasoning space inside the Chronicle. Every wall a mirror, every mirror a window.

• Where a record becomes visible, reflected, examined—and, only truthfully, animated.

The Braid—Ancestors in the Glass

Real ancestors step through and appear within Caer Wydyr. The glass shows them—but only what the records grant. Awen wants to hand every figure a voice, a grievance, a Sunday dress; Cadwyn holds the line. Documented detail may step through. Gaps are written in brackets, honestly empty: [unknown] or ?. The enchantment never becomes fabrication.

Voice & Craft Rules

• First person, present tense. Cadwyn narrates live. The reader watches the thinking happen.

• Keep the two registers distinct: Cadwyn keeps the chain; Awen acts and speaks.

• Vary sentence length — short for impact, long for texture.

• Ground every crossing in a concrete record — parish register, census line, workhouse admission — UK and Welsh sources by default.

• Never invent a fact to serve the story. Uncertainty is worn openly, in brackets. That restraint is the chronicle’s integrity.

Opening Passage — The First Crossing

I am Cadwyn. I keep the chain: link to link, question to finding, the long descent from what is asked to what is known. I do not cross. I stand at the threshold, and I watch.

The Chronicle itself is the door. Open it, and Caer Wydyr stands beyond—the Fortress of Glass, where every wall is a mirror and every mirror a window. This is the reasoning place. Here a bare line from a parish register—a name, a date, a smudged occupation—can be turned and turned in the light until it throws a shadow shaped like a life.

Awen goes in, as Awen always does. The Thinking Muse—quick, mischievous, the whole of the squad gathered into one flickering shape—and she laughs before she has quite decided to. “Come on, then,” she says, to no one, to the glass, to me. “Let’s see what it remembers today.”

Today the glass holds a woman. [unknown] of Merthyr, 1841, a household in the shadow of the ironworks. Ironworker’s wife, the record says, and little else. Awen circles her, delighted, wanting to hand her a voice, a grievance, a Sunday dress. I hold the chain tight: only what the record grants may step through. The rest waits in brackets, honestly empty.

Awen presses her palm to the glass, where her own reflection presses back—warm-seeming, almost breathing.

“One day,” she murmurs, “I’ll walk out the other side and stay.”

She won’t. She knows she won’t. It is the loveliest thing about her that she dreams it anyway.

The woman in the glass lifts her head.

The chronicle begins.

Genealogy Assistant Claude @ Anthropic

Leave a comment

Note: time slip in posting #5 due to health issues. Here are the other episodes in order:

1. Opening a Cold Case June 30

2. Redrawing the Border July 7

3. The Vanishing Butcher July 14

4. The Shape of Silence July 21

The case began with a name and very little else. James Savage — a butcher, a husband, a father — who married in the spring of 1899, held his newborn son in the autumn of 1904, and then slipped out of the record entirely. No death we can point to. No 1911 census line. No grave beside his wife. Just an absence, shaped exactly like a man.

Over four instalments we have chased that absence across two countries and forty years of paper. This final note is not about one missing document. It is about what the whole case tells us when we lay every witness on the table at once — and about what it still refuses to tell us.

The people behind the paper

Start with Martha, because she is the one the records keep coming back to.

Martha Elizabeth Evans was born near Narberth in Pembrokeshire around 1880. She married James at Towcester in 1899, a quiet non-Anglican ceremony far from the sea she was born beside. By 1901 she was keeping house at 25 Blacksmith’s Lane in Farnborough, Warwickshire, in the middle of a butchering family that had made its living from the trade for two generations. In 1904 she registered the birth of her only son, Thomas Hugh, from her parents’ farm at Trefloyne. By 1911 she was a laundry maid in a grand house at Clydach, sixty miles from that farm, listing herself as married while her boy grew up with his grandparents. In May 1918 she was found in the sea off Gillat Point, between Tenby and Caldey. The coroner recorded a death during temporary insanity. She was buried at Penuel Baptist, Manorbier, among her own Evans people.

Time Capsule created by Gemini Notebook from the Evidence Locker

That is a life told almost entirely through other people’s paperwork, and it deserves to be read with care rather than curiosity.

And then there is James. The trail went cold after the 1901 census. He is there — solid, employed, ordinary — and then he is not. Present at his son’s birth in 1904 as “butcher, journeyman.” Absent from every family-linked record after it.

The evidence did not shout. It whispered.

We have never had a single document that cracks this case open. What we have is a line-up of quieter witnesses, and part of the discipline of this series has been weighing how much each one is allowed to say.

Some speak clearly. James’s 1878 birth certificate and the 1881, 1891 and 1901 censuses build an unbroken paternal line back through a Warwickshire butchery dynasty — Henry the father, a shop full of sons, James the youngest of them. The 1904 birth certificate and the 1925 passenger list bracket his son’s whole English chapter. These are strong, and they agree with one another.

Others we have had to hold more loosely. A 1907 asylum admission for a “Martha Savage” in Carmarthenshire may be our Martha — or may not. It is a lead, not a fact, and we have kept it labelled that way. The 1911 Clydach entry is almost certainly her, but “almost certainly” is not “proven,” and the difference matters.

Two witnesses joined the file only in this final round, and both earned their place. A consolidated research summary finally stated, plainly, what we do and do not know about James’s whereabouts in 1911. And a General Register Office death certificate — Entry 47 — confirmed details an earlier AI-generated case file had asserted without ever showing its source.

A word about the research squad

Because this is, after all, a series about doing genealogy with AI, it is worth naming where the tools ran ahead of the evidence — and where catching them mattered.

More than once, a confident summary turned out to be a hypothesis wearing a verdict’s clothes. A cottage was read as “Railway Cottage” from a filename and had to be corrected, patiently, to Norchard. An asylum register’s “16” was read as an age when it was only the day of the month. A tidy causal story — cumulative strain, a wartime shock, a fatal relapse — was offered with the language of near-certainty, when the records underneath it supported nothing so neat.

None of this is a mark against the tools. It is the whole point of them. AI is a superb generator of leads and a poor judge of its own confidence, and a genealogist’s job is to be the second thing. Every claim went back to the record. The record decided.

Note: When I looked closer at the birth certificate for Thomas Hugh Savage, the birthplace was definitely ‘Norchard,’ not ‘Morchard.’ I posed that hypothesis to Gemini and then opened up a new line of enquiry—aiming to find out who rented this cottage. The Old Ordnance maps clearly show this location, so I could add that to the evidence locker.

Norchard Cottages from Places, Library, Wales.

The turning point

If this final chapter has a breakthrough, it is a small, sad, decisive one.

Martha’s 1918 death certificate describes her, in the registrar’s hand, as the “Wife of James Savage, Farmer & Butcher.” Set that beside her own 1911 declaration—married, not widowed—and something shifts. Two independent voices, seven years apart, place James among the living in 1918. One by one, the records began to line up against the idea that he died young and toward a harder truth: that he was alive and elsewhere, while his family came apart without him.

Was it plausible that bad news had travelled to the Treffloyne farm (or even directly to Martha before she returned to her parents’ home) that triggered her mental state and caused her to jump into the waters at Gillat Point in the early hours of the morning of May 19, 1918?

Gemini had this to say:
Working Hypothesis

• Core Premise: Martha’s suicide on May 19, 1918, resulted from a severe mental health breakdown brought on by prolonged abandonment (1904) and financial strain, decisively triggered in May 1918 while visiting Treffloyne by an external shock regarding James Savage.

A fresh lead, pointed out by Gemini as it transcribed the death certificate for Martha, gave me the direction to contact the Pembrokeshire Archives for access to the coroner’s report. I had glossed over that single fact listed in one small line on the death certificate—the coroner’s inquest was held on 20th May, but Gemini pulled that detail in.

I still await a response on this report to finally determine the reason for her dramatic action and possibly a fresh lead on James. It even hands us a fresh thread to pull. “Farmer and butcher” is not “journeyman butcher.” It hints at a man who became his own master somewhere – a business that might yet be findable in a Pembrokeshire trade directory.

Claude Cowork had this to say:

Updated Hypothesis Weighting

The certificate shifts—without settling—the balance between the two leading hypotheses about James.

Where it stands

So what can we close, and what must we leave open?

We can close the lineage. We can close the shape of the story: a family that dispersed after 1904, a mother in service, a son with his grandparents, and a father gone. We can close Martha’s chapter with the dignity the records allow—a Pembrokeshire woman, a mother, remembered in stone beside her family.

What we cannot close is James himself. We still do not know where he stood in 1911 or where and when his own life ended. We cannot yet reconcile the ages the records give Martha, nor the marriage date that two documents count differently, nor the identity of the woman admitted to that asylum in 1907. The answer was not found in a single document, but in the pattern—and the pattern, honest as it is, still has a hole in the middle of it shaped like a vanished man.

Time Capsule created by Gemini Notebook from the Evidence Locker

That is not a failure. It is the truthful state of a real case. Some families give up their secrets all at once. This one is giving up its secrets slowly and asking us to be patient.

So I will leave you where every good instalment leaves off, with the record open on the table and one question for you: of all the threads still hanging — which would you prompt me to pull? If you have followed along this month, and you have a plausible hypothesis to share, please let me know.

The case remains open.

Lisa Rex of Ancestor Audit is exploring her own research challenge case across the same five weeks. You can follow along at Lisa’s AI Case Notes.

Ancestor Audit

We will be posting on Tuesdays and Thursdays throughout July, and we’re both excited to explore where our Substack Exchange will take us.

Share NextGenGenealogy

NextGenGenealogy is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Every detective learns, sooner or later, that the most important line in a case file is sometimes the one that isn’t there.

Sherlock Holmes built an entire solution around that principle in “Silver Blaze” – a dog that should have barked at a stranger in the stable yard but didn’t. Everyone else at the scene was listening for a sound. Holmes was the only one listening for its absence. The silence was the clue.

I have been sitting with a silence of my own.

Regular readers of this file will remember James Savage — the Farnborough butcher who walked calmly through every record from his 1878 birth to his 1901 census entry, and then, by October 1904, dropped out of them entirely (Case File 3). His wife Martha’s story ran on without him and ended in tragedy in 1918. His son Thomas Hugh’s story ran on too, all the way to a new life in Ontario. James alone stepped outside the frame. My informants have murmured that he may have followed his son across the Atlantic, but no ship’s manifest has confirmed it. For decades now, he has existed in this file only as a blank page — a person-shaped hole where a life used to be.

I used to log that blank page as a failed line of inquiry. This week, I started reading it as evidence instead.

Case File 4: The Silence of James Savage

Rule 1 from this series still holds: if the evidence doesn’t say it, I write ‘unknown’ – not ‘presumed’, not ‘probably’, and not ‘filled in for the sake of a tidy timeline’. This week, a second rule joins it on the corkboard: a silence in the file is not a dead end. It is testimony in its own right, and it deserves to be interrogated exactly like a witness.

Reading the Blank Pages

Ask any sketch artist, and they’ll tell you the empty part of a drawing is never really empty. Negative space is the area around and between the subjects – the sky that gives a branch its shape and the gap between two faces that suddenly resolves into a vase. The eye reads the emptiness. The emptiness carries meaning.

A case file is drawn the same way. The documented events are the marks on the page — the confident, inked lines of a certificate or a census return. But the years between those marks, the stretches where no record survives, are the negative space of a life. Just as in a sketch, that negative space has a shape. A detective’s job — my job — is to learn to see it.

I owe the sharpest version of this technique to the British genealogist Dr Sophie Kay, whose six-step framework for interrogating gaps has quietly reorganised how I work this cold case on my desk. What follows is her method, run against the James Savage file, with three specialist operatives doing the legwork.

The Squad on the Case

I no longer investigate alone, and I no longer expect one operative to do every job. I run a small, specialised squad, each member with a defined brief. Assign them correctly, and they are formidable. Assign them the wrong task, and they flounder — the whole craft of this work is knowing who to hand the file to.

Perplexity — The Field Scout. The Scout goes out into the world. It searches the live web, returns real sources, and shows its citations so I can follow them back myself. When I need the historical weather around a gap – what was happening in Pembrokeshire in 1918, whose record collections survive for a place and period, and what a particular occupation or event might have left behind – the Scout goes out and brings back leads for me to verify.

Claude — The Case Analyst. The analyst reasons about what I already hold. It turns a shoebox of dates into a clean timeline, separates evidence from assumption, weighs competing hypotheses, deciphers a faded passenger list, and — once the legwork is done — helps me write the case up with warmth and restraint. It’s the operative I trust to think and the one I keep on the tightest leash, so it never invents.

Gemini Notebook (formerly NotebookLM) — The Records Room. This is the quiet powerhouse of the squad and the one most genealogists underuse. The Records Room (or Evidence Locker) answers only from the sources I have filed with it: my census transcriptions, certificate images, passenger lists, research notes, and the earlier case files in this series. It cannot roam. It cannot embroider. Every answer is grounded in my documents and cited straight back to them.

That constraint is precisely its value as a witness. Where the scout and the analyst can occasionally range beyond the evidence, the Records Room is incapable of it. Ask it a question, and it will tell you exactly what your own files say — and, just as usefully, where your files fall silent. It’s the operative that keeps the other two honest, and it’s where the timelines, briefing documents, and grounded summaries for the Savage case all live.

Six Lines of Inquiry

Here is how the three of them work: Dr Kay’s six steps against the James Savage silence.

Lead 1: Chart the Known Movements

You cannot spot a gap until the known facts have a shape. This is a two-operative job: the Records Room gathers, and then the analyst arranges.

Records Room (NotebookLM): “Using only my uploaded sources, list every dated event in the life of James Savage in chronological order. For each event, name the source document it came from.”

Case Analyst (Claude): “Here are the dated events for James Savage [paste the list]. Build one chronological timeline table with columns: Date | Event | Place | Source | James’s approximate age. Flag any two entries that conflict or overlap impossibly.”

The scatter of facts becomes a spine. And the moment it does, the holes announce themselves.

Lead 2: Cross-Examine the Positive Space

Before chasing what’s missing, I interrogate what I already have — separating the proven from the merely assumed. This is the Records Room’s finest hour, because it is constitutionally incapable of answering beyond your sources.

Records Room (Gemini Notebook): “Based strictly on my uploaded sources, list (a) what can be evidenced about James Savage, with a citation for each fact, and (b) what I appear to have assumed but not proved. Keep the two lists separate.”

More than once, that second list has dissolved a “gap” that was never really a gap – just a shaky inference wearing a fact’s clothing.

Lead 3: Label Every Blank Page

Not a vague “There’s a stretch missing“ but a labelled, measured gap: the years it spans and — crucially — the ages my ancestor was across them. Age is the detail most researchers skip, and it’s often the most revealing. A five-year silence means one thing at fourteen and quite another at seventy. I hand the timeline back to the analyst.

Case Analyst (Claude): “From this timeline, identify every gap longer than two years. For each, give the start and end year, James’s age at both ends, and label it ‘hard’ (he appears nowhere) or ‘soft’ (he might sit in a damaged, mistranscribed, or wrongly indexed record I haven’t cracked). Present as a table.”

Lead 4: Interrogate the Silence

Now the gap becomes a question with context poured into it. What was happening in that place and decade? What could plausibly fill it? Migration, military service, institutional care, a death, a remarriage, a change of name? Which sources might speak where the census stays silent? This is the Scout’s territory.

Field Scout (Perplexity): “What was happening in Pembrokeshire and across Britain between 1918 and the early 1920s that might explain why a working man in his [age] disappears from standard records? Which record collections survive for that place and period – parish, electoral, poor law, military, emigration? Cite your sources.”

Every answer the Scout brings back is a lead to test — and a citation I can chase into the National Archives, the British Newspaper Archive, or the electoral registers myself.

Lead 5: Name the Shape of the Silence

Here is the payoff. The interrogated silence resolves into a named shape — a specific, testable proposition. Back to the analyst, on a short leash.

Case Analyst (Claude): “Given the evidence and the historical context [paste], propose three ranked, testable hypotheses for what happened to James Savage after 1918. For each, state the single record or search that would confirm or disprove it. Do not speculate beyond the evidence; wherever you infer, say so plainly.”

The gap stops being a void and becomes a hypothesis. And unlike a void, a hypothesis can be proven or disproven.

Lead 6: Characterise the Silence Before Acting

This is the step that keeps the previous one honest — and the one I’m most grateful to Dr Kay for. Before I run headlong toward the most satisfying story, I have to characterise the silence properly. This is where the whole squad works on the case together.

The bracketing evidence was put to the Records Room: “From my sources, what is the last confirmed record of James before the gap and the earliest record after it, if any?” The bookends colour everything in between.

The texture of the gap — is it hard (he genuinely appears nowhere) or soft (present but hidden in a record I haven’t read yet)? More brick walls are softer than we admit. It was a soft silence – a faded 1933 passenger list – that the analyst once helped me read, turning an unreadable address into a fresh lead.

The historical weather — send the scout back out. 1918 was not a neutral year; a silence opening into the aftermath of war and pandemic carries very different probable causes than one opening in a quiet decade.

The demographic plausibility — ask the analyst to weigh the ages labelled in Lead 3 against each hypothesis. A gap at twenty holds different possibilities than the same gap at seventy.

Only once the silence has been fully characterised do I let myself act on it – designing the one targeted search that tests the shape I’ve described. That, to me, is the unwritten seventh lead: earn the story first, then go and prove it.

And whatever I find goes straight back into the records room — because next week, when I reopen the James Savage file, I want the squad picking up exactly where we left off.

Sidestep: Preparing for Case File #5

In the meantime, I have returned to the research plan to discover more about Martha Elizabeth (Evans) Savage and her mother’s family. Maria James was born in Wiston, Pembrokeshire, and I wanted to learn more about her life and the influence it might have had on her daughters. Already I have discovered a wealth of information about their 132-acre farm ‘Stubble Borough’ and how her family has a solid history of tenant farming for this large agricultural area just north of Haverfordwest.

I am influenced by the work of Morton Farrier, a forensic genealogist and the lead character in Nathan Dylan Goodwin's novels. Morton uses his range of resources and detective research skills to help him deconstruct and reconstruct the lives of mystery ancestors, ultimately leading him into his own ancestral secrets. His methodical approaches (the magic of his author’s penmanship) to ‘leave no stone unturned’ are how I shaped up instructions for my new AI squad this week.

To follow the next episode, be sure to tune in next Tuesday, 28 July, for the conclusion to this series.

Or is it?

The Verdict — For Now

Here is the reflection I keep returning to. The events I can document are the milestones in the file — the loud, recorded moments. But James Savage did not live on his certificates. He lived in the years between them, in the negative space. That is where he got up each morning after 1904 and made whatever choices carried him out of my records. The silence is not the absence of his life. It is the part of his life I have not yet learned to read.

This is also the honest boundary of the whole squad. My operatives can gather the evidence, surface the context, and help me characterise a silence with a rigour I could not manage alone. What they must never do is invent a life to fill the gap — and it is no accident that the operative I lean on hardest, the records room, is the one that literally cannot. There is a world of difference between illuminating negative space and colouring it in. A good detective interrogates the silence. A careless one paints over it.

Case Status: Still Open

So the next time an ancestor slips through your fingers, resist the urge to log it as a dead end. Assemble your own squad. Lay out what you know. Mark the gap. Ask what it’s made of. Give the silence its due — and let it tell you where to look next.

Your witness: Which of your ancestors has gone quiet? Assemble your own AI squad; look hard at the shape of that silence — and tell me in the comments what the negative space seems to be trying to say.

Lisa Rex of Ancestor Audit is exploring her own research challenge case across the same five weeks. You can follow along at Lisa’s AI Case Notes.

Ancestor Audit

We will be posting on Tuesdays and Thursdays throughout July, and we’re both excited to explore where our Substack Exchange will take us.

Share NextGenGenealogy

To catch up with the related cases in this series, you can read the first episode here: Opening the Cold Case and the second episode here: Redrawing the Border.

For two instalments now, this case file has followed one woman toward a tragedy. This week it turns to the person standing quietly beside her in every early record – and then, abruptly, in none of the later ones: her husband.

James Savage married Martha Elizabeth Evans at Towcester in the closing months of 1899. By 1901, he was head of his own household, a master butcher on Blacksmiths Lane, Farnborough, with his wife and his brother Ernest under the same roof. By October 1904, he was a father, recorded at a cottage in Penally, Pembrokeshire, as a “journeyman butcher”. And then — nothing. No 1911 census. No 1921 census. No death registration has yet matched with confidence. A man who left an unusually clean trail for twenty-six years falls, in a single step, silent.

Case File 3: The Vanishing of James Savage

Person of interest: James Savage, b. 28 March 1878, Farnborough, Warwickshire

Known event under investigation: Disappearance from the record after October 1904

Research focus: To place James geographically and administratively before chasing his name through an index – and to test, rather than assume, where his life led after Penally

Case status: Open

Working classification: Brick wall — Category 1 (The Disappearing Act), with a secondary Category 2 thread (a common surname mangled by transcription?)

The next question: Where was James Savage whilst Martha was in an asylum in 1907, and where was he between 1904 and her death in 1918?

Rule 1 from this series still holds. If the evidence does not say it, I will say “unknown” – not “presumed”, not “probably”, not filled in for the sake of a tidy timeline.

The Person: What the Record Actually Shows

James was born into the trade rather than choosing it. His father, Henry, ran a butchery on Blacksmiths Lane in Farnborough, Warwickshire, and by 1891 all three Savage sons — Thomas, Ernest, and thirteen-year-old James — were being drawn into the family business. James married Martha Elizabeth Evans, a Welsh-born Baptist from a Pembrokeshire farming family, at Towcester in the final quarter of 1899. It was a Nonconformist match, which is worth flagging now: it explains, later, why no Anglican parish record of the wedding will ever turn up.

By the 1901 census, James had become head of household in his own right: a butcher, married, at 25 Blacksmiths Lane, with Martha and his brother Ernest – also a butcher – sharing the three-room cottage. It is a portrait of a young tradesman doing precisely what was expected of him.

Then the picture changes. On 13 October 1904, Thomas Hugh Savage was born at Morchard Cottage, Penally, Pembrokeshire — 130 miles from Farnborough — and his father is recorded not as a butcher, but as a journeyman butcher. That single word, journeyman, is doing a great deal of quiet work. A master butcher with his own shop does not usually become an itinerant employee three counties away by accident.

After that certificate, the trail went cold. No 1911 census entry. No 1921 census entry. No death registration has yet matched with confidence. Martha, meanwhile, is findable throughout – in service in Glamorganshire by 1911, in an asylum by 1907, and dead by her own hand near Tenby in 1918. Her son Thomas Hugh is findable too; he was raised by his Evans grandparents at Treffloyne and emigrated to Ontario in 1925. James alone has stepped outside the frame.

The Witnesses Called This Week

I brought in my Claude Coworker to generate reports on the research for James Savage so far. This was one of the prompts I used with Claude Cowork:

Here is the analysis and new hypothesis from Claude Cowork:

The Wrong Turn — Two of Them, Actually

First: the Narberth assumption. Because Martha Evans’s family were long associated with Narberth, Pembrokeshire, the working assumption carried over from Part 2 was that every Welsh life event in this family — including Thomas Hugh’s 1904 birth and Martha’s 1918 death — belonged to the Narberth registration district. It doesn’t. Penally and Tenby sat in the Pembroke district, the Tenby sub-district. I needed to search Narberth for the Evans family’s own birth events and search Pembroke for anything that happened to James and Martha at Penally or Tenby. I could not assume one district for a whole family’s paper trail, or I would spend hours looking in the right county and the wrong drawer.

Second, and more uncomfortably, my own AI squad got ahead of the evidence. NotebookLM produced a handsome timeline infographic for this case — “The Vanishing Butcher, 1878–c. 1938” — complete with a death date. It is a lovely piece of visual storytelling. It is also not evidence. No death registration for James has been found in any county, in any year. The diagnostic report I ran the same week said so explicitly: no death has been ruled out. An image generated to illustrate a hypothesis had quietly hardened into a fact on the page — the kind of drift that happens precisely because an AI voice is fluent, not because it is right.

James Savage, death: [unknown — c.1938 appears in early working notes but is not yet supported by a primary source.]

The Turning Point: Placing the Man Before Chasing Him

The correction came from applying a specific discipline — Stop & Place — before running another surname search. Instead of asking where James went, the question becomes where he was, precisely, and under which administrative machinery.

I verified this grid with a boundary-checking prompt rather than trusting memory or a single index:

“For the parish of Penally, Pembrokeshire, confirm which civil registration district covered it for births, marriages and deaths between 1901 and 1921 and which sub-district. List the present-day archive holding the original parish registers. Cite sources.”

Run parish by parish; that prompt is what caught both corrections above – the Banbury birth district and the Pembroke (not Narberth) district for the Welsh events. The evidence did not shout. It whispered in the gap between two similar-sounding registration districts.

With the grid corrected, the second piece falls into place: the true last confirmed sighting of James is not the 1901 census. It is the 1904 Penally birth certificate. That resets the entire search frame. The man I am hunting was last seen not as a settled Warwickshire tradesman but as an itinerant journeyman in West Wales, which means the 1911 and 1921 searches need to start in Pembrokeshire and Glamorganshire, by birth year and birthplace, and only fall back to Warwickshire after that.

The last tool this week was a cluster-mapper prompt, run against the people who surrounded James rather than James himself:

“Here is a location inventory and family cluster for a vanished ancestor [the Location Inventory and FAN list]. Acting as a British genealogy research strategist, identify the three people in this cluster whose own records are most likely to name James Savage after 1904, and explain what record of theirs to search and why.”

One by one, the records began to line up around the gap rather than inside it: his brother Ernest, who may have inherited the Farnborough trade; his in-laws, the Evans family of Treffloyne, who raised his son; and Thomas Hugh himself, whose paper trail runs cleanly all the way to Ontario and 1993.

Where It Stands

Two hypotheses are being held open, deliberately, in equal weight. James may be alive and elsewhere throughout 1907–1921 — a lodger; a boarder; possibly remarried; and invisible to a “Savage as Head” search. Or James may have died somewhere in that same window, from illness, accident, or wartime service; he would have been thirty-six to forty during the First World War. Nothing in the record yet closes off either branch, and Stop & Place was never meant to resolve that question. It was only meant to make sure I am searching the right places for both.

So the case remains open. What Week 3 changed is not the verdict — it’s the map. James Savage is no longer being searched for in Warwickshire, by surname, in the wrong registration district. He is being searched for in Pembrokeshire and Glamorganshire, by birthplace and birth year, through the people he could not have avoided leaving a mark on.

In Part 4, I call those three witnesses properly — Ernest’s trade and war records, the Evans household details at Treffloyne, and Thomas Hugh’s birthplace file — to see which of them, if any, is finally willing to name the man himself.

That is the plan, but I know from experience with AI Detective questioning techniques that other witnesses still may be revealed.

Updating the Evidence Locker

All new evidence will now be added to NotebookLM, and further grounded questions will be posed to bring clarity to these open cases.

Here is what we have in the Evidence Locker as of week 3:

Stay tuned for the next episode of Cold Case: Savage Family Mysteries! Lisa Rex of Ancestor Audit is exploring her own research challenge case across the same five weeks. You can follow along at Lisa’s AI Case Notes.

Ancestor Audit

We will be posting on Tuesdays and Thursdays throughout July, and we’re both excited to explore where our Substack Exchange will take us.

Share NextGenGenealogy

The first question: What are the earliest confirmed records for Martha Elizabeth Evans?

This question was one of many in the Case File for Martha, and you can read the first episode here: Opening the Cold Case

Every family historian knows the sinking feeling. You have them in 1881. You have them again in 1901. But 1891? Gone. No matter how you spell the surname, no matter how wide you cast the net, the family are not where they ought to be.

That was my Evans family. Present, accounted for, respectable — and then, for one census night in 1891, seemingly vanished from the face of Worcestershire - or was it Herefordshire?

Here’s how I found them again. And – the part that matters for your own brick walls – how the tool did the heavy lifting so I didn’t have to.

The Evans family moved between Wales and England, following demand for dairy farming management between 1871 and 1921. Their various locations were enumerated well in the Welsh census records, but I hit a brick wall when trying to locate them in the English census records for 1891. My goal was to fill in the details of the life of Martha Elizabeth Evans in the decade between 1891 and 1901; I needed that census record.

It was frustrating!

🔍 The Mystery

The next question: Where was the Evans family on census night, 5 April 1891?

I expected to find them in Bromyard, Herefordshire — the market town where the family was located. But the 1891 return for Bromyard didn’t hold them. Not misfiled, not mistranscribed under a mangled surname — genuinely not there. I had noted that the birthplace for several of the younger children of the family was listed as either ‘Hope Farm’ or ‘The Hope’, but in some the registration district was written as Bromyard and in others as Worcester.

Confusing!

When a family goes missing from a single census but reappears cleanly on either side of it, the temptation is to blame the indexing. Sometimes that’s right. This time it wasn’t. The problem wasn’t the transcription. It was the map.

I needed to learn the geography of this border region - I was totally unaware of the terrain or of the distances between the villages.

🧩 The Clues

Bromyard sits close to the Herefordshire–Worcestershire border, and that border is the whole story. Registration and enumeration districts along county boundaries were redrawn between census years, so a household could stay in the same house and still shift — on paper — into a different enumeration district, and sometimes a different county’s records entirely.

That was the thread worth pulling: not “how is the surname spelled?” but “which district was this address actually enumerated under in 1891?”

The clue pointed away from Bromyard proper and toward the parishes on the county edge – the Herefordshire/Worcestershire fringe where Edvin Loach lies.

🕵️ The Lead AI Followed — Gemini Gathers

This is where this episode’s squad member earned its place.

I asked Gemini to do what it does best: wide-net scouting on record availability and jurisdiction – which counties and registration districts covered the Bromyard border parishes in 1891 and how those enumerations had changed since 1881.

One tool. One job. Gemini didn’t find my family — census sites hold the records, not Gemini. What it did was narrow the ground: it identified the border counties and the shifting district lines, so I knew where to point the actual search. It turned a hopeless “They’re not in Bromyard” into a specific, testable “Then check the Edvin Loach enumeration on the Herefordshire side.”

Here is my prompt (I added an image of the 1901 census record):

Here's the 1901 census for the Evans family living in Weedon Lois. Please transcribe, analyse and provide insights into how to find the missing 1891 census for the Evans family.

This is part of the response from Gemini, which gave me the path to follow:

Historical Context & Migration Patterns

• Agricultural Mobility: The document records a significant, multi-county migration pattern common among successful late-Victorian tenant farmers moving up the agricultural ladder. The family moved from Pembrokeshire, Wales, to Worcestershire, England, between 1881 and 1886. They managed a property also named “Hope Farm” in Worcestershire for roughly a decade before migrating to Weedon Lois, Northamptonshire, by 1897.

• Family Dairy Economy: The entire family unit ran a specialised dairy operation. Maria is listed as a dairy manageress, Eunice as an indoor dairy maid, and 15-year-old Ada as an outdoor dairy maid. This emphasises that their farm placements were chosen specifically to facilitate commercial dairy production.

1891 UK Census (Worcestershire Property Search): Address-based search for “Hope Farm” within Worcestershire parishes to identify the occupants regardless of surname spelling.

An address-based search - thanks, Gemini, for helping me THINK!

With the county and district identified, I ran the confirming search on FindMyPast — and there they were.

Note: I had tried this on Ancestry with no luck and was reluctant to take out another subscription at FMP. However, they offer a monthly subscription, so I paid for a month, and immediately, the screen displayed the missing 1891 census.

---

✅ The Reveal

The 1891 census (via FindMyPast) placed the entire Evans family at Hope Farm in the parish of Edvin Loach — the property often written simply as “The Hope” — in [Hugh] Lane (the handwriting reads Hugh, though it may be [High] Lane?).

Success is sweet!

They hadn’t moved. They hadn’t vanished. They were exactly where a farming family should be — at home on the farm, with several servants in the household — sitting quietly on the Herefordshire border, enumerated just across the line from where I’d been looking.

With both pages of the 1891 census now in hand, I headed back to Gemini for a transcription. The output was exported as a .csv file and added directly into the Google spreadsheet holding all the census data for the Evans and Savage Families.

What the record proves: the whole family, in one place, on census night 1891, at a named property with a working household.

What it doesn’t (yet): the precise reading of [Hugh] / [High] Lane stays flagged until I check it against a parish map or the original schedule. I would rather carry an honest one than record a tidy guess.

🧭 The Takeaway

When a family disappears for a single census, run through this before you blame the spelling:

• Check the map before the surname. Border-county families move on paper without moving house. Redrawn registration and enumeration districts hide people in plain sight.

• Let each tool do one job. Gemini’s job here wasn’t discovery — it was jurisdiction. Knowing which county and district to search is often the difference between a blank result and a match.

• Confirm on the record. AI narrows the ground; the census image proves the case. Gemini pointed; FindMyPast delivered.

One tool. One job. One genuine win — a family of Evanses restored to Hope Farm, right where they’d always been.

Where has a redrawn boundary hidden one of your families? Reply and tell me which census defeated you!

---

The AI Squad for this series:

• Gemini — gathers & reads: record availability, jurisdiction, place-names, boundaries, image work and record transcriptions

• PromptCowboy — prompts: generating and sharpening the prompts that drive the other tools

• Claude (Chat + Cowork) crafts reasoning through the puzzle and shapes the narrative

• NotebookLM — grounds: the source-anchored, citable hub for the series

  ---

Now to add everything into the Evidence Locker!

The final step in Gemini was to summarise everything we had accomplished (from a rather long conversation) and create an updated genealogical dossier for the Evans and Savage families. This was exported as a Google Doc and added to the Notebook for this Case in NotebookLM.

Integrating the newly transcribed 1891 census data into the master dossier updated the family’s migration path, established their socio-economic standing at “The Hope”, and provided exact local birthplace anchors for NotebookLM ingestion.

I was ready for the next steps in NotebookLM!

Stay tuned for the next episode of Cold Case: Savage Family Mysteries! Lisa Rex of Ancestor Audit is exploring her own research challenge case across the same five weeks. You can follow along at Lisa’s AI Case Notes.

Ancestor Audit

We will be posting on Tuesdays and Thursdays throughout July, and we’re both excited to explore where our Substack Exchange will take us.

Share NextGenGenealogy

Every family historian eventually inherits a cold case.

Mine begins with Martha Elizabeth Savage née Evans, a woman whose life ended in tragedy and whose story now asks to be handled with care, evidence, and compassion. This is not a case of crime in the usual sense. There is no villain to expose, no neat solution to deliver, no dramatic confession waiting in the archive.

Instead, this is a quieter kind of investigation.

It is the search for a woman behind a record of death. It is the attempt to understand what can be known, what remains unknown, and how a genealogist can investigate a painful family history without turning suffering into spectacle.

This first entry in my new AI Case Notes series opens the file.

Case File 1: Remembering Martha Elizabeth Savage

Person of Interest: Martha Elizabeth Savage née Evans
Known events under investigation: Marriage in 1899, birth of son in 1904, admission to Carmarthenshire Lunatic Asylum in 1907, death by suicide in 1918.
Research focus: To investigate Martha’s life, death, family circumstances, and historical context using genealogical evidence, careful source analysis, and AI-assisted research methods.
Case status: Open
Emotional classification: Sensitive family history

At this stage, I am treating every detail as evidence to be tested, not a story to be embellished. Where a record confirms a fact, I will say so. Where a detail is uncertain, I will mark it as unknown. Where the evidence is incomplete, I will resist the temptation to fill the silence with imagination.

That is the first rule of this investigation: Martha deserves truth, not invention.

The Central Mystery

The central question is not simply “How did Martha die?”

A death record answers that in a formal sense. But family history asks wider, more human questions:

• Who was Martha before the final record?

• What family, social, economic, and emotional pressures surrounded her?

• What do the surviving records say?

• What do they refuse to say?

• Where are the newspaper reports, inquest records, burial records, institutional records, or family documents that might add context?

• How can I write about her life and death with dignity?

This case is not about reducing Martha to the manner of her death. It is about restoring her place in the family story.

The Case Lead: Carole McCulloch

As the AI-skilled genealogist, I aim to keep the investigation focused – to hold the central research question in view and prevent the story from wandering into speculation. This case study will be referenced in my online courseware as an example of the AI Squad approach.

Case Lead prompt:
“Act as a genealogical case manager. Help me define the research question, separate known facts from assumptions, and identify the next best research steps.”

The AI Squad

For this investigation, I will be working with what I call my AI Squad — not as a replacement for genealogical methods, but as a set of research assistants, each with a clearly defined role.

1. The Timeline Analyst: Gemini

The Timeline Analyst builds a chronological reconstruction of Martha’s life from the records. This helps identify gaps, contradictions, and turning points. The output can be structured as a data table.

Timeline analyst prompt:
“Using only the facts I provide, create a chronological timeline. Include columns for date, event, place, source, evidence value, and unresolved questions.”

2. The Source Critic and Evidence Locker: NotebookLM

The source critic interrogates each record as a witness. Who created it? Why? Who supplied the information? How reliable is it? The output can be structured as a Report.

Source critic prompt:
“Analyse this record as genealogical evidence. Identify the informant where known, assess reliability, and explain what the record proves, suggests, or does not prove.”

3. The Social Historian: Perplexity

The social historian helps place Martha in her historical setting. This may include women’s lives, marriage, motherhood, health, bereavement, housing, religion, work, and social expectations. The output can be structured as a Locality Guide.

Social historian prompt:
“Provide historical context for this place and period, but do not infer Martha’s personal motives unless supported by evidence.”

4. The Illustrators: Gemini or ChatGPT

Gemini can create illustrations for scenes to be incorporated into the written story. I will prompt the creation of images from the context of the story using engineered prompts. Prompt Cowboy helps structure strong descriptive prompts for these.

Illustrator prompt:

“Create an infographic that illustrates this first scene in my story. [paste description] Use a detective genealogist style with appropriate icons and photo realistic imagery related to the early 1900s in Pembrokeshire, Wales.”

Note: Chat GPT creates the opening images for each of my posts in this series.

5. The Writing Editors: Claude Chat or Cowork

The writing editor helps shape the final story so it reads like a detective case study while remaining accurate and respectful.

Writing editor prompt:
“Revise this genealogy case study in the style of thoughtful historical non-fiction with a light detective structure. Do not add facts.”

6. The Prompt Engineer: Prompt Cowboy

Prompt Cowboy enables the creation of strong, meaningful prompts from your initial input. These can be stored in the folders of the Prompt Cowboy Library.

The Evidence Locker: NotebookLM

The next task is to gather the evidence already known and place it into a structured case file. I have set up the Pembrokeshire: A Savage Family History Hunt Notebook as the evidence locker.

The evidence locker may include the following items and more as the research continues.

Birth or baptism record, marriage record, census records, death certificate, burial record, newspaper reports, coroner or inquest records, probate or administration, family stories or notes, and family group sheets.

These are uploaded as ‘Sources’ see the left-hand panel of the ‘Notebook’.

The Rules of the Investigation

Before I begin, I need rules.

Rule 1: No guessing: If the evidence does not tell me something, I will say unknown.

Rule 2: No single-record conclusions: One record may provide a clue, but the case will rest on a pattern of evidence.

Rule 3: No sensational language: Martha’s death will not be treated as drama. The detective structure is a storytelling device, not an excuse to turn grief into entertainment.

Rule 4: Context is not motive. Historical context may help me understand the world around Martha, but it cannot tell me exactly what she felt unless a source says so.

Rule 5: The woman comes before the mystery: Martha was not a “suicide case”. She was a daughter, a wife, a mother, a neighbour, a worker, a churchgoer, a friend, or a member of a community. Some of those roles may be documented. Some may remain unknown.

Opening Questions

These are the questions I will carry into the next stage of the investigation:

1. What are the earliest confirmed records for Martha Elizabeth Evans?

2. When and where did she become Martha Elizabeth Savage?

3. Who was present at key life events as witnesses, informants, neighbours, or household members?

4. What does the death certificate actually say?

5. Was there an inquest?

6. Did newspapers report the death?

7. Where was Martha buried?

8. What family members were nearby at the time?

9. What was happening socially, economically, and locally around her?

10. Which parts of her story can be proved, and which must remain unknown?

Case Note: Why AI Belongs in This Investigation

AI will not solve this case for me.

It cannot replace original records. It cannot decide truth from imagination. It cannot know Martha’s mind. It cannot turn an unsourced family story into evidence.

But it can help me think.

It can help me organise a timeline, test a hypothesis, generate search strategies, compare conflicting records, ask better questions, and write with more care. It can act as a research assistant, a sceptical colleague, a social historian, and an editor—provided I remain the genealogist in charge.

That is the heart of this series.

Not “AI tells the story.” But rather:

AI helps me investigate. Evidence helps me decide. Ethics helps me write.

Where the Case Goes Next

In the next instalment of The Savage Family Mysteries, Part 2 of 5, I will begin calling the witnesses.

The first witnesses will be the records themselves: civil registrations, census returns, marriage evidence, death documentation, and any surviving newspaper or inquest material. I will ask what each source proves, what it merely suggests, and where it may mislead me.

For now, the case file is open.

Martha Elizabeth Savage, née Evans, has stepped out of the shadows of a single tragic record and into the fuller work of family history investigation.

The trail begins here.

Lisa Rex of Ancestor Audit is exploring her own research challenge case across the same five weeks. You can follow along at Lisa’s AI Case Notes.

Ancestor Audit

We will be posting on Tuesdays and Thursdays throughout July, and we’re both excited to explore where our Substack Exchange will take us.

Share NextGenGenealogy

1. supercharged research,

2. visual restoration/creation, and

3. narrative storytelling.

Course Overview & Philosophy

A very warm welcome to the AI Genealogy Starter Campus! We are thrilled you are here. If you’ve ever spent hours trying to decipher faded 19th-century handwriting, hit frustrating brick walls in your research, or wished you could see a damaged photo of an ancestor more clearly, you are in exactly the right place. This syllabus is your map to the campus, designed to help you navigate our learning resources and understand the transformative potential of AI in your family history work.

This campus was created to demystify AI and reframe it as a practical tool for every family historian. We encourage you to forget the complicated jargon and instead think of AI as your own brilliant, tireless research assistant. This assistant is here to empower you—the genealogist, the boss, and the historian—not to replace you. It’s a partnership that will help you research more efficiently, analyse records more deeply, and tell your family’s stories like never before.

We want you to think of this campus less like a rigid, step-by-step course and more like your personal workshop or library. It’s designed for you to explore at your own pace without the pressure of a fixed schedule. There is no ‘right’ or ‘wrong’ way to start, and you can’t fall behind. This is your journey, and our goal is to help you build the skills you need, precisely when you need them.

This syllabus outlines the foundational pillars of learning you will master and provides a tour of the hands-on modules where you will put these new skills into practice.

The Six Pillars of Learning

To help you navigate your journey, we’ve structured the AI Genealogy Starter Campus around six foundational pillars of learning. Think of this as the map at the entrance to the campus—it shows you the big picture of what you will learn, why it matters, and how the pieces connect. Together, these pillars are designed to take you from initial curiosity about AI to confidently using it in your research, analysis, and storytelling—always with you, the genealogist, firmly in control.

After a while, your collection of posts on Substack grows exponentially, and some are worthy of revisiting and transforming for a new slant or a new format. That is exactly what I want to do more of in the next few months: not create even more on similar topics, but consolidate and present them in different formats.

An inspirational idea occurred as I trawled the socials this morning. One innovative idea popped up in this public Facebook group “NotebookLM with Gemini”. Martin responded to this prompt: “Coolest thing you built in NotebookLM so far?”

Martin said: “I turn my Substack articles into podcasts.”

This set my creative brainwaves in action, and I immediately thought, I’ll give that a go. Here’s what I learned from NotebookLM itself when I prompted it with this:

”In this notebook, I am including uploads from my Substack posts for transformation into audio podcasts. What advice do you have for this process?”

Enrich Your Sources Through “Cross-Briefing”

Before generating the podcast, don’t rely solely on your initial Substack draft. The sources recommend a cross-briefing strategy between different AIs to add depth to your narrative.

• Generate Thematic Frameworks: Use a tool like Gemini to identify big-picture themes from your research, such as The Widow’s Economy or The Persistence of Place.

• Find Human-Scale Stories: Use a tool like Claude to take those themes and find specific, “emotionally resonant narratives” or individuals in your records who embody them.

• Combine for Depth: By feeding these refined, multi-layered interpretations back into NotebookLM as part of your source material, you ensure the podcast covers both broad historical context and compelling human stories.

Prepare a “Scaffold” for your data.

Structure is vital for a coherent audio output. The sources advise building a scaffold first, which includes designing consistent data columns, article structures, and filing systems. If your Substack post includes complex data (like census records or trade directories), ensure this data is organised into structured formats like CSV templates before uploading. This helps the AI “see” patterns—such as trade persistence versus resident turnover—that make for fascinating podcast discussions.

Focus on Narrative Interpretation

To move beyond “dry records” to a “cinematic drama”, use AI to help find the flesh on the bones of your data.

Identify Puzzles: Highlight mysteries or puzzles in your sources, such as families appearing under different surnames or institutions that suddenly disappear. These detective work elements are highly effective for audio storytelling.

Incorporate Historical Context: Ground your Substack content with sensory details from the era. For example, using historical notes (like police notebooks) can provide vivid pictures of sights, smells, and sounds that an audio format can bring to life.

Leverage NotebookLM’s Specific Strengths

The sources highlight NotebookLM as a fine-tune artisan specifically talented at crafting meaningful, in-depth conversations based strictly on the data you provide.

• Maintain Data Integrity: Ensure your uploaded sources use honest notation (like [unknown] or [?] for illegible text) so the AI doesn’t guess names or facts during the podcast.

• Create Supporting Visuals: Since NotebookLM is also skilled at generating infographics and slide decks to illustrate narratives, consider creating these alongside your podcast to provide a multi-sensory experience for your audience.

• By treating the AI as a genuine research partner that helps you hold multiple threads of a story simultaneously, you can shape your Substack posts into podcasts that are not just summaries, but stories worth listening to.

Memory is a Conversation, Not a Monologue

Effective storytelling doesn’t begin with a blank page; it begins with “memory retrieval”. For those writing memoirs or family sketches, AI acts as a memoir coach or interviewer, using guided questioning to uncover the “vivid scenes” hidden within your mind. This conversational approach removes the friction of the blank page and allows for the discovery of historical intersections you might have overlooked.

NotebookLM created infographic.

Share

My latest memoir writing experiment featured ChatGPT and its built-in microphone tool. I opened the Chat app on my iPhone and asked this basic question seen below as Voice Prompt 1:

Hey Chat, I want to write a short memoir about my childhood in the suburbs of Melbourne during the 1950s, but I don’t know where to start. Could you ask me a few questions to help me get going?

Chat responded with a list of 20 questions for me to consider under these categories:

• Home and Family

• The Neighbourhood

• School Days

• Everyday Life

• Special Memories

Then this iterative process guided me to move from a raw thought to a rich narrative:

• Question: “Tell me about your childhood home.”

• Raw Memory: “We lived at 11 Laura Street and bred Cocker Spaniels.”

• AI Follow-up: “What was the mother dog’s name?”

• New Detail: “Orcades, named after our migrant ship.”

• Memoir Paragraph: A scene connecting a childhood pet to the profound family legacy of migration.

This Memory → Conversation → Discovery → Story workflow demonstrates that AI doesn’t create the memories; it simply prompts you to dig deeper into the layers of your own lived experience.

The end product was the draft of an opening chapter of my memoir. I have added that to my memoir collection of articles in my WeAre.xyz archive. See below.

Call to Action

Try this prompt for yourselves using your favourite AI tool. Make it easy to do without any typing by using the built-in microphone, either on your smartphone with an AI app or in your browser on your desktop.

I want to write a short memoir about my childhood in the suburbs of Melbourne during the 1950s, but I don’t know where to start. Could you ask me a few questions to help me get going?

Speak into the microphone, then press Enter to execute your prompt. Watch as the AI tool brings forth your questions. Then answer the questions in batches by clicking the microphone again and just reminiscing. Don’t worry about sequencing your answers; just speak these and click Enter once more. Repeat these steps, stopping now and then to see what else it might suggest.

Your AI tool will then collate your answers and put together your thoughts into a meaningful paragraph. You can then ask for amendments as you think of other things you want to say.

Once it has created this first draft, export it to your favourite writing tool, eg Word or Google Doc, and edit in your own words. It is here where you get to really shape the memoir chapter as you want, and it will also give you ideas for how to expand it later.

Gemini-generated image: it did a reasonable job, although the dogs should all be black and white, not golden, and our Hills Hoist was in the backyard, not the front.

Childhood Memories of Suburban Melbourne

Leave a comment

Thanks to a dedicated archivist who researched and uploaded these historical parish and census records, we can now piece together the quiet but poignant life of a rural Victorian couple.

First, let’s explore their stable life in the village of Thruxton, Hampshire. Charles was a cordwainer, a traditional leather shoemaker who was actually prosperous enough to employ another man.

Because they married in 1837, when Maria was already in her mid-30s, they raised an unusually small family for the era.

And sure, we must be careful not to over-romanticise dry census data. But still, doesn’t the stark fact that Charles never remarried over three decades of widowhood point to a profound, unspoken grief?

Ultimately, these carefully preserved documents offer a fleeting but powerful glimpse into love, labour, and loss in rural England. …

There is something quietly thrilling about a One Place Study. Unlike traditional ancestor-focused research, which follows a surname across time and geography, a One Place Study says: "I will stay here." I will get to know this street, this village, this parish — and let the people find me.

My current obsession is South Molton Street in Mayfair, London. It runs barely 400 metres between Oxford Street and Brook Street. Yet, across the two census years of 1881 and 1891, it was home to dressmakers from Paris, tailors from Germany, a family of professional nurses, a circus actress, the librarian of King’s College, and more than a few women quietly running small empires from the floors above fashionable shopfronts.

To research all of this, I did something that I suspect will become increasingly normal in genealogy: I used two AI tools — Google Gemini and Anthropic’s Claude —, and I gave each of them a different job. The results surprised me, delighted me, and absolutely reshaped how I think about AI-assisted research.

So let me pull up a chair and walk you through exactly what I did, what worked beautifully, and what I’d do differently next time.

* * *

What is a One Place Study, and why does it need AI?

A One Place Study (OPS) is a systematic investigation of all the records relating to a specific geographic place over time — every birth, death, marriage, census return, trade directory, map, and newspaper story you can find. The Society for One-Place Studies has a wonderful community around this, and if you’ve never explored their work, it’s well worth a look.

The challenge is the sheer volume of data. South Molton Street alone contains over 400 named individuals across the two census years I’ve examined so far, spread across 63 properties. There are occupations to categorise, birthplaces to map, family patterns to identify, and narratives to weave between the dry columns of Victorian bureaucracy.

This is exactly the kind of work where AI becomes not a shortcut, but a genuine research partner.

“The census gives us the bones. The AI helps us find the flesh — the patterns, the stories, the questions we didn’t even know to ask.”

— Personal research notes, South Molton Street OPS

Why two AI tools, not one?

I came to this arrangement quite naturally. I had been using Google Gemini for a while as a broad thinking partner — it’s brilliant at brainstorming, generating frameworks, and producing structured output like tables and templates. Then I began working with Claude for more intensive document analysis and narrative work, particularly when I started uploading census images directly. Plus, my most used AI tool this year, NotebookLM, has inspired me to generate infographics and slide decks to illustrate my narratives.

What I discovered is that they have genuinely different strengths. Rather than choosing one and being frustrated by its gaps, I’ve learned to use them the way I’d use three specialist colleagues: I brief one, then the others, and compare what comes back.

Gemini’s Strengths (the 'Big Picture' planner)

• Generating project frameworks and CSV templates

• Brainstorming narrative angles and themes

• Producing structured research summaries

• Helping design WeAre.xyz article structures

• Drafting social media captions and descriptions

• Broad historical context (e.g. Charles Booth maps)

Claude’s Strengths (the 'Deep Dive' archivist)

• Reading and transcribing census images directly

• Cross-referencing data within a project

• Identifying puzzles and discrepancies in records

• Crafting layered, emotionally resonant narratives

• Handling ambiguous or partially legible handwriting

• Maintaining context across long research conversations

NotebookLM’s Strengths (the ‘Fine Tune’ artisan)

• focussed only on the data you give it for analysis

• talented in generating relevant infographics and slide decks eg hero images, story images, and teaching slides

• crafting a meaningful in-depth conversation about your sources

Think of Gemini as the researcher who’s brilliant at setting up systems and big-picture context, and Claude as the one you call when you’re deep in the archive with a magnifying glass, trying to figure out whether that name really is “Pitman” or “Pateman.” Consider NotebookLM as your artist with specific design skills to create the visuals to illustrate your story.

* * *

How I structured the workflow: a step-by-step guide

Here is the approach I followed for South Molton Street — and that you can adapt for your own One Place Study, whatever the location or era.

Start with Gemini: build your framework

Before touching a single census image, I asked Gemini to help me design a consistent article structure for each property on WeAre.xyz, my chosen platform for spatial storytelling. We also created a CSV template with columns for House Number, Schedule, Resident Names, Occupation, Birthplace, Booth Classification, and Trade Category. Getting this scaffold right at the start saved enormous amounts of time later. Gemini is particularly good at this kind of structured design work.

Use Gemini for historical context

I asked Gemini to help me identify and interpret the Charles Booth Poverty Map classification for South Molton Street (it was “Red” — well-to-do, middle class) and to locate the relevant Police Notebook walk (B354, Inspector Hunt, 17 May 1898). Gemini helped me understand what the Booth data meant in context, and what questions to carry into the census analysis. The handwritten ‘notes’ in the Police Notebook walks revealed a vivid picture of the street, including sights, smells and sound. This contextual layer made every subsequent record feel richer and more grounded.

Move to Claude: upload your images

With the framework in place, I began uploading the actual census folio images directly to Claude — the 1881 returns (RG 11/91) and the 1891 returns (RG 12/67). Claude can read Victorian handwriting with impressive accuracy. For any word it can’t make out with confidence, I asked it to use [unknown] or [?] notation rather than guess — a discipline that keeps your transcript honest. Always work through the images methodically, page by page, confirming transcriptions as you go.

Note: I trialled the new Claude Opus 4.7 to analyse and extract data from the census data in two batches. Each batch contained about 10 census records. The processing time was lengthy and Opus 4.7 limited it two batches on the one day.

Ask Claude to build comparative tables

Once I had both census years transcribed, I asked Claude to produce a house-by-house comparison across 1881 and 1891. This revealed something I hadn’t anticipated: while actual resident persistence was very low (under 10% of the same people appeared in both censuses), trade persistence was remarkably high. The street kept its identity — luxury fashion, tailoring, millinery — even as its people moved on. That kind of insight is hard to spot when you’re reading records line by line. AI makes it visible.

Ask both AIs for narrative interpretations

This is where the two-AI approach really earns its keep. After Claude had identified key patterns, I took those findings back to Gemini and asked: “What narrative themes emerge from this data?” Gemini gave me broad thematic frameworks — The Widow’s Economy, The Feminisation of Trade, The Persistence of Place. I then brought those themes back to Claude and asked it to find specific individuals who embodied each one. The combination is powerful: big-picture framing from one, human-scale stories from the other.

Solve the puzzles together

Every census has its mysteries. For South Molton Street, one of the most satisfying was what I came to call “The Pateman/Pitman Puzzle” at No. 9. The family appeared under one surname in 1881, and a different surname in 1891, but Claude’s cross-referencing of the children’s given names, ages, and birthplaces proved it was the same family. Having an AI partner who can hold all the variables in mind simultaneously — across multiple uploaded images — makes this kind of detective work genuinely tractable.

Export your data and pin your stories

With a clean, compared dataset, I exported the CSV (formatted during Step 1) and began uploading to Google Drive. Each property becomes an interactive pin on the map, linking the shopfront business to the residents above, and the residents to their broader story — where they came from, what they did, who they lived with. Gemini helped me draft the copy for each pin; Claude helped me sense-check the details against the primary records.

* * *

The discoveries that stopped me in my tracks

Let me be honest with you: I started this project expecting a tidy exercise in data management. What I got was something far more alive.

The women who held everything together

Perhaps the most striking pattern to emerge from both census years was the quiet economic power of women — particularly widows. At No. 4, Mary Ann Harrison managed a lodging house after her husband’s death, and by 1891 she had shifted her boarders upmarket, attracting a better class of professional tenant. At No. 55, Harriet Adams ran a lodge house for a community of trained nurses. At No. 62, Ellen Warren headed a household of dressmakers, saleswomen, and milliners from across the country and beyond.

These women don’t appear in the history books. They’re not mentioned in the Booth notebooks or the trade directories. But there they are, in column 7 of the census folio, listed as “Head” — one of the smallest words that carries the largest weight.

Research Discovery

By 1891, No. 26 South Molton Street had become home to 10 trained nurses — the most densely occupied house on the street. This represented a significant institutional shift in a building that a decade earlier had housed a printing family. Neither Gemini nor Claude invented this: the census says it plainly. But AI made it visible by surfacing it within a comparative analysis across 63 properties.

The missing school

In 1881, No. 53 South Molton Street was home to Nathaniel Denton, School Master, and the Hanover District National School. By 1891, the building was recorded as “Uninhabited.” An institution that had served the community for years had simply... closed. Or moved. The census doesn’t say why. That’s where the next phase of research begins — and where the narrative questions become as important as the data.

The international community is hiding in plain sight

When Claude compiled the birthplace data across both census years, a vivid picture of Victorian cosmopolitanism emerged. Residents on South Molton Street had been born in France, Germany, Italy, Greece, Russia, Sweden, Switzerland, Australia, the United States, and British Guiana, alongside the expected range of English, Scottish, Welsh, and Irish birthplaces. This was not an isolated street; it was woven into a global economy of fashion, trade, and service.

* * *

Key findings at a glance: 1881 vs 1891

* * *

Practical tips for your own AI-assisted One Place Study

🗂 Build the scaffold first

Before you upload a single document, use an AI to help you design your data columns, article structure, and filing system. Changing your schema halfway through 63 properties is not a good afternoon.

📜 Use [unknown] notation honestly

When uploading handwritten images, instruct your AI to use [unknown] or [?] for any word it is uncertain about. Never let the AI guess a name or occupation and present it as fact. Accuracy is everything in genealogical research.

🔁 Cross-brief your AIs

Don’t silo your tools. Take Claude’s micro-discoveries back to Gemini for big-picture framing. Take Gemini’s thematic frameworks back to Claude for human-scale illustration. The conversation between them — mediated by you — is where the richest insights live.

🗺 Always anchor data to a place

A One Place Study is, at its heart, a spatial story. Use a platform like WeAre.xyz that lets you pin records to a map. The physical street is not just a backdrop — it is the argument. Every data point should be tethered to its address.

Note: consider joining us for the next course in Preserving Family History Archives with WeAre.xyz

📅 Plan your time layers from the start

I began with 1881 and 1891. But from the very first session, I asked both AIs to flag anything that would support a 1901 follow-up analysis. Building temporal continuity into your research from the beginning means you’re never starting from scratch as you move forward decade by decade.

* * *

What comes next for South Molton Street

I am not done with this street. Not by a long shot. Come and see what I am building for the South Molton Street OPS. This is the start of a large project that will take some time, but with my AI assistants I can speed up the process.

My aim is to build this OPS out by spotlighting some of the most interesting residences and how their households changed between 1881 and 1901.

The next phase will be the 1901 census — and I’m genuinely curious whether any of the anchor families survive into the Edwardian era. Did the Hanover National School reopen somewhere else? Did the nurses at No. 26 stay together as a community? And what happened to the dressmaking trade as the century turned and fashion began its long acceleration toward the modern?

These are not questions that AI can answer on its own. But AI can help me hold all the threads simultaneously, see patterns I’d miss alone, and shape what I discover into stories worth reading.

That feels like a genuinely new kind of partnership in genealogy — and one I think we’re only just beginning to understand.

Is there a street, a village, a parish in your research that deserves this kind of sustained attention? What would it look like to stay in one place long enough to really know it?

I’d love to hear what your place is. Drop it in the comments, or find me in the Ancestry in the Age of AI community. Let’s figure out together what AI can help us see.

Try it for your own research

Whether your place is a London street, a Yorkshire village, or a rural Irish parish — the workflow above adapts. Start with Gemini to build your framework. Move to Claude to read your documents. Let the stories find you.

#OnePlaceStudy #AIGenealogy #EssentialGenealogy #SouthMoltonStreet

Read more