VaultBook for Genealogists: A Private Offline Family History Research Vault for DNA Notes, Sources and Family Photos
Margaret Ellison — Meg to everyone who has ever sat with her at a courthouse microfilm reader — did not set out to keep secrets. For fourteen years, since she retired as a school librarian in Ohio and finally had the hours her younger self never had, she had done what careful genealogists are taught to do. She built the tree properly. She cited her sources. She attached census pages and draft cards and marriage returns to the right people in the right places, and she synced her desktop genealogy software to her online tree so cousins could see the work and, occasionally, correct it. Her file held about twenty-four hundred people: four generations deep in Ohio and Pennsylvania coal country on her mother’s side, an Irish line that arrived with little more than a parish name and a stubborn oral tradition, a German line that was meticulous on paper until it suddenly was not. The tree was the public record of the family, as far as she could prove it. It was also, she discovered on a Tuesday evening in October while reviewing DNA matches she had been putting off for months, no longer where the real work could safely live.
The match itself was not dramatic in the way television makes it dramatic. There was no single revelation, no stranger’s name that explained everything. There was a cluster. Three matches, related to each other, sharing far more DNA with Meg and with her first cousin than any explanation through the man on her tree — the man who had raised her maternal grandfather, whose surname her grandfather carried, whose photograph sat in a frame her mother had dusted every week — could account for. Meg did what librarians do. She did not gasp and rewrite the tree. She started a spreadsheet, then abandoned the spreadsheet, then pulled the birth record again, then the delayed birth certificate, then the baptismal entry from a church register she had photographed in Pennsylvania three summers earlier and never fully transcribed. The ages did not line up the way she had assumed they lined up. A witness name she had read as a neighbour now read, in this new light, as something else entirely. By midnight she had a working hypothesis that was strong, well-evidenced, and nowhere near proven: the man who raised her grandfather was almost certainly not his biological father. Someone else — a man whose name appeared twice in the records, at the edges, never in the centre — very likely was.
That distinction, between strong and proven, is where genealogy actually happens, and it is also where Meg’s system broke. She could not put this in the tree. The tree synced. The tree was shared. Cousins she barely knew copied from it, as cousins do, and within a month a careful hypothesis written in a note field would appear on four other trees as established fact, stripped of every qualifier she had attached to it. She also could not leave it in a spiral notebook on her desk, which is where, if she was honest, the most sensitive page of notes from that night ended up. She needed a working vault, and she needed it to be built for the way this work actually fails.
The Four Stresses That Break a Genealogist’s Note System
Every experienced researcher has a version of Meg’s week. Ask in any genealogy forum, any society meeting, any late-night thread where people confess how they actually work rather than how the manuals say they should, and the same four stresses surface again and again, in different words, attached to different software. They are not edge cases. They are the ordinary weather of serious family history work, and a note system that cannot withstand them will not fail loudly. It will fail quietly, over years, in lost context and copied errors and record images that turn out never to have been yours at all.
The first stress is custody, and it usually arrives disguised as a billing event. A researcher pays for a records subscription for years — many serious researchers pay for more than one — and builds diligently inside it. Every census household, every probate packet, every newspaper clipping gets attached to the tree with a satisfying click. The tree fills up. It looks owned. It feels like an archive. Then life intervenes: a price rise that is hard to justify in a lean year, a household decision to cut subscriptions for a while, a simple choice to let one membership lapse because the current project is in county records anyway and the subscription can wait. What the researcher discovers, usually at the worst moment — mid-argument, checking a detail she was sure she had — is that much of what she “saved” was never saved at all. She saved a link. The image of the record lives on the provider’s servers, behind the paywall, and when the membership lapses the attachment remains visible as a reference while the image itself goes dark. The citation survives. The evidence does not. For work that may be picked up and put down across decades, that is not a minor inconvenience. It is a structural flaw in the archive.
The companion frustration is how hard the platforms make it to do the right thing in advance. Downloading your own research, record by record, is technically possible and practically punishing. Images come down one at a time, through a viewer, with no sensible bulk path. They arrive with opaque numeric filenames that mean nothing three months later — a string of digits that could be a death certificate or a city directory page — and the context that made the image meaningful, which person it belongs to, which fact it supports, which other record it contradicts, lives in the tree, not in the file. So the diligent researcher ends up with two bad options. She can spend hours renaming and re-filing downloads she may never need, a clerical project that competes directly with research time. Or she can trust the attachments and move on, which most people, most of the time, rationally do. Meg had done both, in patches, across fourteen years, which is why her laptop held a downloads folder of several hundred record images in which perhaps a third were named well enough to trust, a third could be identified by opening them, and a third were, functionally, lost — present as files, absent as evidence. Community discussions return to this point with a particular weariness, because everyone in them has the folder. The folder is universal. So is the quiet knowledge that the work inside the subscription is rented access wearing the clothes of ownership.
The second stress is sync, and anyone who keeps a tree in two places — desktop software as the working copy, an online tree as the sharing copy, or the reverse — knows its particular dread. Sync is supposed to be plumbing: invisible, boring, reliable. In practice, researchers describe a ritual of hesitation before pressing the button. The connection hangs. The program demands a sign-in it demanded last week and the week before, as if the account were a door that relocks itself out of habit. A sync that should take minutes stretches, stalls, restarts. Worse than the stalls are the warnings. The two versions diverge, not in one dramatic fork but in a hundred small edits made on whichever side was open that day, until no one — including the researcher herself — can say with confidence which tree is current, and reconciling them becomes a project so unappealing that it joins the downloads folder in the category of things to be dealt with later.
For Meg, sync divergence had a sharper edge than staleness. Her desktop tree and her online tree were not identical in what they contained, and the difference was deliberate in some places and accidental in others, and after fourteen years she could no longer reliably reconstruct which was which. A note she believed she had kept desktop-only: had it travelled in the last sync she ran in the spring, before she understood what the warning screen meant and clicked through it because dinner was ready and the alternative was not syncing at all? She genuinely did not know. That uncertainty is the real damage sync trouble does to a researcher working with sensitive material. It is not only that the plumbing is annoying. It is that the boundary between private and shared — the boundary her whole ethical handling of the DNA finding depended on — ran through a process she did not fully trust and could not fully audit. A system whose privacy depends on remembering what a sync dialog said in April is not a private system. It is a hope.
The third stress follows directly, and it is the one researchers discuss most carefully, because it involves other people’s feelings and other people’s copying habits. Genealogy is collaborative by nature and promiscuous by default. Online trees are built to be copied: a leaf, a hint, a one-click acceptance, and another researcher’s ancestor — or error — becomes yours. This is wonderful for the well-documented lines, the census-rich, certificate-backed stretches where copying mostly propagates truth. It is corrosive for everything still in progress. A speculative line — the possible father, the tentative parish, the immigrant ancestor who might be yours or might be a same-name stranger two townships over — placed in a shared tree does not stay speculative. Other researchers copy the person without copying the reasoning. The qualifier, if it exists at all, lives in a note field that does not travel with the copy. Within a season, the hypothesis has hardened across a dozen trees into apparent consensus, and the original researcher finds her own tentative idea cited back to her as independent confirmation. Every serious genealogist has been burned by this in one direction or the other: either their speculation escaped and fossilised, or they built six months of work on someone else’s speculation that arrived dressed as fact.
DNA work has raised the stakes of this stress beyond anything the older etiquette anticipated. A tree note about a possible misattributed parentage is not like a note about a possible birthplace. It concerns living people — Meg’s mother had living siblings; the cousin whose DNA made the cluster visible was very much alive and had tested, originally, to learn about Irish heritage for a family reunion booklet. It concerns people who did not consent to have the question asked, let alone published. It can be wrong: DNA evidence is powerful and probabilistic, matches can mislead when pedigrees collapse or when a community intermarried for generations, as coal-country families did, and a written accusation in a shared tree has a permanence that a private worksheet does not. And it concerns timing. Even if the hypothesis proves true, the researcher may need months of documentary work, and possibly difficult family conversations, before anything belongs in a shared space. What she needs in the meantime is a place where the work can be as thorough as the public work — logged, sourced, argued — while being private by architecture rather than by settings. A tree platform’s privacy controls are settings: toggles that can be misread, defaults that can change, a living-person filter that protects the recent generations while leaving the sensitive note attached to a deceased grandfather visible to anyone with tree access. Meg’s finding sat exactly in that gap. Her grandfather was long deceased. The platform considered him public. The finding was anything but.
The fourth stress is the quietest and, in the long run, the most costly: single fragile copies, and context that lives separately from the thing it explains. Every long research project accumulates a class of material that exists once. The interview recording on a phone, made in a kitchen in 2019, never backed up because it was going to be transcribed next month. The photograph identification that lives in the researcher’s memory — she knows the woman in the second row is Great-Aunt Rose because her mother told her in 2014 — never written down because, at the time, it felt unforgettable. The letter from a second cousin in Pennsylvania, scanned once, the original since lost in a house move, the scan itself in an email account’s attachments under a subject line no one would think to search. It retires, it forgets, it — in the case of a ninety-two-year-old aunt who is the last person alive who can name the tall boy at the edge of the picnic photograph — it has a closing date that nobody chose.
Meg’s shoebox made this concrete in a way that a decade of good intentions had not. Of the sixty photographs, she could identify perhaps half the subjects with confidence, from her own knowledge. Another quarter she could identify with her aunt’s help, if the help was asked for soon and asked for properly — not “who is this?” over the phone, but a structured session, photograph by photograph, with the answers recorded, dated, and attributed: identified by whom, when, with what degree of certainty, and what story came with the identification. The remaining quarter might never be identified, and even that needed recording, because a documented “unknown, shown to two family members in 2026, neither recognised the setting” is research value — it stops the next researcher, possibly Meg herself in five years, from starting at zero. What the shoebox needed was not a scanner, though it needed that too. It needed a record system in which the photograph, the identification, the identifier, the date, and the uncertainty all lived together, permanently attached to each other, in more than one place. Her existing system attached identifications to nothing. Her mother had pencilled names on backs of photographs and the pencil was fading. Her aunt’s memory was the backup copy of the pencil. That is not sentimentality about old photographs. That is a preservation emergency with faces in it.
Hold the four stresses together and a pattern emerges that explains why so many capable researchers feel vaguely behind in their own projects. The tree is good at being a tree. It was never designed to be a custody system for evidence you must own outright, a reliable boundary between draft and publication, a log of work in progress, or a home for singular, context-dependent material. Asking it to be those things produces the folder of numeric downloads, the feared sync button, the escaped speculation, the shoebox. Meg did not need a better tree. She needed the other half of the system: a private working vault, designed around the stresses rather than in denial of them, where everything the tree could not hold could be kept with the same rigour she brought to the tree itself.
What the Working Vault Has to Hold
The easiest way to understand Meg’s vault is to walk its shelves, because the inventory is the argument. Nothing in it is exotic. Every category exists in every serious researcher’s project already — in notebooks, folders, email threads, camera rolls, and memory. The vault’s contribution is not inventing new kinds of material. It is giving each kind a permanent, private, findable home where it can be connected to the others, so that a question asked in 2029 can be answered from work done in 2026 without archaeology.
The foundation is the research log, one per ancestor or per research question, and it is the habit whose absence researchers regret most consistently. A research log records the work, not just the results: on this date, I searched this collection, for this person, under these name variants, with this outcome — including, especially, the negative outcomes. The empty search is evidence. The parish register checked and found silent for the relevant years, the census index searched under four spellings with no plausible candidate, the deed books for a county examined for a decade in which the family supposedly bought land and did not: these dead ends are what stop a researcher from re-walking the same ground every eighteen months, which is what happens, universally, when they are not written down. Meg’s logs live in her vault as one entry per research session, labelled by the ancestor and the question, linked to the entries for the people and places involved. That is the entire point of it.
Beside the logs sit the proof arguments, and these are the vault’s most unmistakably genealogical contents. A proof argument is the written case for a conclusion that no single record states outright: that this Mary in the 1870 census is the same Mary who married in 1874; that the German immigrant in the naturalisation index is, despite the mangled surname, the man in the family Bible. Professional standards in genealogy have insisted for decades that such conclusions be written out as arguments — evidence assembled, conflicts resolved, reasoning exposed — rather than asserted as facts with a citation or two attached. In practice, most hobbyist trees contain the conclusions and almost none of the arguments, because trees have nowhere to put an argument. It attaches, awkwardly, as a note, truncated and unformatted, or it lives in a word-processing document on one computer, named something like Smith_line_notes_FINAL2. They are the work, filed where they can be found, revised with their history kept, and read alongside the evidence they interpret.
The record images are the vault’s answer to the custody stress, and Meg’s rule for them is absolute: an image she relies on gets downloaded, renamed in plain language, and filed in the vault at the moment she first uses it, while the context is fresh. Not next month. Not in a batch. The 1880 census page for her coal-country great-grandparents is in the vault under a name a human can read, attached to an entry that says whose household it shows, which lines matter, what the enumerator got wrong about the children’s ages, and which other entries cite the same page. The baptismal photograph from the Pennsylvania church register — the one that mattered so much on DNA night — is there too, with its transcription, her translation notes for the Latin, and a remark about the witness whose role she now reads differently. This discipline changes the meaning of her subscription entirely. She describes it as the last time she will ever have to do it, which, at sixty-one, with the tree at twenty-four hundred people and growing, is a calculation about the rest of her researching life.
Photo identification records form their own category, distinct from the photographs themselves, and the distinction is the lesson of the shoebox. Each significant photograph in Meg’s vault has an entry of its own: the image, a description of the physical original and where it resides, the people identified in it, position by position, and — critically — the provenance of each identification. Identified by her aunt, on a recorded date, with high confidence, with a story attached about the picnic and the church that hosted it: that is one grade of evidence. Identified by Meg herself from resemblance to a labelled portrait: a lower grade, recorded as such. Conjectured from a uniform and a date range: lower still, and labelled as conjecture in the entry’s first line, where no future reader can miss it. The aunt sessions became the vault’s most urgent project in the weeks after the shoebox arrived. Genealogists of all people should recognise the difference; it is the difference between a source and a rumour, applied to their own families’ photographs.
Interviews occupy the next shelf, and not only the aunt’s. Over fourteen years Meg had accumulated the usual miscellany: a recorded conversation with her mother in 2016, made on an early phone and rescued twice from dying devices; a call with the Pennsylvania second cousin, recorded with permission, in which he read aloud from a family register in his possession — a register Meg has never seen, whose current whereabouts after his move to assisted living are, she realised while building the vault, something she needed to establish and record; two shorter recordings from a family reunion, valuable mostly for one story about the Irish line’s arrival that contradicts the written tradition in an interesting way. In the vault, each recording is an entry with the audio attached, a full transcript, a summary of the genealogically relevant claims indexed by the people they concern, and a note of the circumstances: who was present, what prompted the story, what the speaker could and could not have witnessed personally. That last discipline matters more than it looks. Now it exists as evidence in a system.
The DNA working notes are the shelf Meg built the vault for, and they are its most sensitive contents by a wide margin. Here live the match analyses: clusters charted and re-charted, shared-match tables copied while a subscription tool was available, calculations of how much DNA the known relationships predict and how much the matches actually show, the list of hypotheses ranked by fit, the documentary checks each hypothesis demands, the results as they come in. Here lives the working file on her grandfather’s parentage — the midnight notes from October, cleaned up the next morning but not sanitised, because a working file that hides its author’s reasoning is useless; the evidence for, the evidence against, the alternative explanations she has not yet eliminated, including the one she hopes is true and the one she fears is. Here live the correspondence notes about testing itself: which cousins have tested, who manages their kits, what each cousin has and has not consented to share, who must be spoken to before any conclusion moves anywhere. And here live the things that will never be published in any form: the match who is plainly the child of a secret no one in the family has spoken aloud, recognised by Meg from the pattern and never to be raised by her; the relative who asked, explicitly, that her test results not be discussed beyond Meg. Meg keeps hers in the one place in her system designed to hold things that must not travel.
Correspondence logs and deadline tracking complete the inventory, and they are the shelves a librarian would build first. Genealogy runs on other people’s replies: the county clerk whose office holds the deed books, the parish secretary who will search a register when time allows, the archive in Ireland that acknowledges requests in batches, the historical society with one volunteer who knows the coal company records. Meg’s correspondence entries log every request — what was asked, of whom, on what date, with what fee, what was promised, what arrived, what did not — linked to the ancestor entries the request concerns. The practical payoff arrived within a month: a register search she was certain she had requested in the spring turned out, on inspection of the log, never to have been sent, only drafted; the draft was still in her email, and the log’s honest gap saved her from writing a peevish follow-up about a delay that was hers. It can, at least, refuse to let a deadline pass unnoticed because it was written on a sticky note.
What the vault deliberately does not hold is the tree itself, and Meg is insistent on this point because it is the one her genealogy friends question first. The relationships — the structured web of people, unions, events, and citations that genealogy software exists to manage — stay in the genealogy software, synced to the online tree for sharing, exactly as before. The tree program is her system of record for the tree because it is genuinely good at the tree’s own problems: consistency checks that flag a child born before a parent, duplicate detection, relationship calculation, report and chart generation, the disciplined data model that keeps twenty-four hundred people coherent. The vault is not a rival tree and does not try to be one. It is the research layer around the tree: everything that precedes a conclusion, everything that supports it without fitting inside it, everything that must accompany it privately, and everything that must never follow it into public. The division is not a workaround for two imperfect tools. It is a recognition that a published tree and a private research practice are different kinds of object, with different rules, and that fourteen years of strain came from asking one system to be both. The tree says what Meg can stand over. The vault holds how she got there, what she is still unsure of, and what she knows but cannot say. A complete genealogist needs both sentences to be true at once, in different rooms of the same house.
Why the Vault Is a Single Offline File
Knowing what the vault must hold is only half the design problem. The other half is what kind of object the vault itself should be, and Meg’s four stresses answer that question more precisely than any feature comparison could. The custody stress demands that the evidence be a copy she possesses, not a link she rents. The sync stress demands that the private boundary not depend on a sync process behaving itself. The speculation stress demands privacy that is architectural — true because of how the thing is built, not because a setting was left in the right position. The fragility stress demands that singular material and its context live together, in a form that can be backed up whole, moved whole, and opened decades from now without asking any company’s permission. VaultBook meets those demands in a specific and somewhat unfashionable way: it is a single-file HTML application, local-first, that makes zero network requests. The entire vault — the software and the data together — is one file, opened in a browser, worked on locally, saved locally. Each property of that design maps onto a stress, and it is worth tracing the mapping carefully, because for this use case the properties are not conveniences. They are the difference between a system that can be trusted with a family’s unproven truths and one that merely promises to be trusted.
Start with the boundary, because for Meg it came first. A conventional notes application, however good, is usually a window onto somewhere else: an account, a server, a sync service that decides what lives where. Privacy in such systems is real but derivative — it holds because the provider’s controls work, because the sharing settings were understood correctly, because nothing was ever dragged into the wrong notebook or attached to the wrong shared page. VaultBook’s privacy runs the other way. The application makes zero network requests. There is no account to sign into, no server holding a copy, no sync that could carry a private entry somewhere public while its owner is making dinner. Meg’s DNA working notes are not private because a toggle says so. They are private because the file they live in sits on her machines and goes only where she deliberately carries it. It is the technology’s chief virtue.
The internal structure of the vault is built from entries, labels, and wiki links, and for genealogy this turns out to be a surprisingly exact fit — closer, in daily use, than the folder hierarchies researchers usually impose on their material. An entry is one unit of work: one research log session, one proof argument, one photograph’s identification record, one interview, one DNA analysis. Labels sort entries the way a researcher actually thinks — by surname, by line, by county, by status, so that everything marked with the Irish line and the status of unresolved can be gathered in one view regardless of what kind of material it is. Wiki links do the connective work that makes the vault a system rather than a pile: Meg’s proof argument about her grandfather links to the log entries that record the searches, to the entries holding the record images, to the DNA analysis, to the correspondence entry for the parish request still outstanding. Follow the links and you walk the whole case, in any direction, from any starting point. Nothing sits in isolation unless she chooses to leave it there.
For the most sensitive material, the vault adds per-entry encryption: individual entries can be encrypted with AES-256-GCM, with the key derived through PBKDF2, using the Web Crypto capabilities built into the browser itself. The granularity is what matters for this use. Meg does not need her whole vault locked to the same degree — the research log for a well-documented 1880 census household is not a secret, and treating it as one would only make daily work slower. The DNA shelves are another matter. The grandfather analysis, the unpublishable match, the cousin’s testing correspondence: these entries are encrypted individually, so that even if the file itself were opened by someone else — a borrowed laptop, a repaired machine, a copy sent to the wrong place — the entries that could hurt people remain sealed behind their passphrase, while the ordinary work of the vault stays ordinarily accessible. This is the practical meaning of privacy by architecture, at the level of a single researcher’s habits. Not a vault door on the whole building, though the building is already private by design, but a locked drawer for the papers that need one, inside a house that is not on any public street. It also disciplines the researcher, gently. Deciding which entries to encrypt is deciding, explicitly, what the sensitive material actually is — an exercise Meg recommends to any genealogist who has never sorted their notes by “what would I not want read aloud at a reunion.” The answer is always a smaller and more specific pile than expected, and knowing its exact contents is itself a form of care.
A research vault that cannot change its mind safely will, over years, teach its owner not to change her mind at all, and genealogy cannot afford that lesson. VaultBook keeps version history for entries, and Meg uses it the way she once used the margins of her notebooks, except that the margins are now complete. When the Pennsylvania baptismal entry’s witness changed meaning on DNA night, she did not overwrite her original transcription notes; the entry grew a new layer of interpretation, and the old reading — neighbour, unremarkable, noted in 2023 and dismissed — remains in the history, dated, with her reasoning at the time. That preservation is not nostalgia. In genealogy, a superseded interpretation is evidence about the research process, and occasionally it turns out to have been right for reasons the researcher later forgot. Proof arguments in particular live by revision: Meg’s German surname argument exists in its current form because its earlier forms were safe to improve. Version history is how the inheritance stays honest.
Search is where the vault repays its building costs most visibly, because Meg’s questions are almost never “open the entry I am thinking of” and almost always “find everything that touches this.” VaultBook’s search covers the full text of entries, and it reaches into the material researchers usually cannot search at all: optical character recognition, using tesseract, reads the text inside her record images, so that a photographed register page or a scanned newspaper clipping becomes findable by the words printed on it; attachment indexing extends the same reach to the files she has filed; and a semantic question-and-answer layer lets her ask the vault a question in plain language and be pointed to the entries that answer it, across phrasing differences that would defeat a simple keyword hunt. Consider what this does to a real research moment. A new DNA match names an unfamiliar surname from the coal country. In the old system, Meg would search her tree for the surname, find nothing, and stop — while the surname sat unrecognised in three places she could not grep: a witness line in a photographed baptism, a fellow passenger in a transcribed interview story, a boarder in a census household she had downloaded and never fully read. In the vault, one search crosses all of it, because the images have been read, the transcripts are text, and the entries are linked. It makes the whole layer answer questions at once, including questions its owner did not know, when she filed the material, that she would ever ask.
The last pieces of the design sound like mere convenience until they are mapped onto a genealogist’s calendar, at which point they become load-bearing. VaultBook includes a built-in calendar and a sidebar organised around what is recent, what is due, and what is expiring. Meg’s deadlines — the archive fee that must be paid before a request lapses, the parish office’s narrow window for register searches, the subscription she plans to work hard in its final month and then release — live in the same file as the research they belong to, attached to the entries that explain them, rather than in a separate planner that knows the date but not the reason. There is the vault, carried whole, like the research bag she has hauled for fourteen years — except that this bag holds several hundred entries, every record image she relies on, every transcript, every argument, and it weighs nothing, and its most sensitive pages are locked, and nothing inside it is waiting for a sync to decide where it belongs.
None of this makes VaultBook a genealogy program, and Meg would be the first to object if it claimed to be one. It does not draw her charts, check her tree for consistency, or calculate her relationships; the tree software still does all of that, and does it well, and her online tree still shows her cousins the proven shape of the family. What VaultBook is, in her system, is the answer to the four stresses — the place built for the work the tree was never built to hold. Custody, because the evidence in it is a file she possesses outright, whose record images she filed herself and can open on the worst day of any subscription year. Privacy, because the working layer makes no network requests, answers to no account, and locks its most sensitive entries individually, so speculation and DNA secrets stay working papers until she — not a sync dialog, not a cousin’s one-click copy — decides they are ready to be anything else. Continuity, because singular things and their context live together in entries that carry their provenance with them, in a single file that can be backed up whole and opened with nothing more exotic than a browser, by her, or in time by whoever inherits the twenty-four hundred people and deserves to inherit the reasoning too.
The Research Log: One Entry Per Ancestor, One Honest Trail
Meg Ellison does not start a research day by opening her tree. She starts it by opening the research log entry for the one person she intends to work on, and reading the last three lines she wrote to herself. That habit, more than any database she has used in fourteen years, is what stopped her repeating work she had already done.
Every ancestor who is more than a name and two dates gets a dedicated log entry in VaultBook. Not a note stuck to a profile in the tree program — the tree stays the system of record for concluded relationships and sourced events — but a working entry of her own, titled with the person’s name, approximate years, and the place that anchors them. Inside, the structure never varies, and the lack of variation is the point. At the top: the current objective, stated as a question that can actually be answered. Not “find out about Thomas” but “determine whether the Thomas Reilly in the 1900 census in Scranton is the same man who signed a marriage licence in Wilkes-Barre in 1897.” Below the objective, a dated list of repositories searched. Below that, the part most hobbyists skip and professionals never do: the negative searches, written out in full. Below that, findings, each tied to the record that produced it. At the bottom, next steps, in the order she intends to take them.
The negative search is where this log earns its keep. In genealogy a search that finds nothing is not a failed search. It is evidence, but only if you can prove later that you actually performed it, where, and how thoroughly. Meg writes it the way she once wrote catalogue searches as a school librarian: repository, collection, date range covered, search terms and variants used, and the result. “Searched St. Mary’s parish baptismal register, microfilm held at the diocesan archive, for Reilly, Riley, and O’Reilly baptisms, October 1888 to December 1892 — no Thomas, no child of a Thomas.” Six months later, when a cousin emails a tree that places a baptism squarely in those years, she does not have to wonder whether she missed it or simply never looked. She looked. The entry says when, and how, and under which spellings.
Labels do the sorting that folders never could. She labels every log entry three ways at once: by surname, by place, and by record type worked. A single entry can carry Reilly, Lackawanna County, and Parish Registers, and sit comfortably under all three when she filters. When a research trip — real or virtual — is organised by place rather than by person, she pulls every entry labelled with that county and works the list top to bottom, so one visit to a courthouse index or one session in a digitised deed book serves nine ancestors instead of one. When a new record set comes online, she does the reverse: filter by record type, find every ancestor whose log says that type is still unsearched for them, and work that list. The log stops being a diary and becomes a queue.
Wiki links turn the separate entries into a web she can walk. An ancestor’s log links to the entry for the place he lived, and the place entry — hers are small gazetteers of her own making, with boundary changes, parish histories, mine openings and closures, and which records survive for which years — links back to every ancestor she has placed there. It links sideways, too: a log entry for a man links to his brother’s log, his employer’s entry if she keeps one, the church whose register she is working through. Following those links before she starts is her reality check. If the place entry says the parish was only founded in 1894, she will not waste a morning hunting an 1891 baptism in its register, however confidently a copied tree claims one. The links also surface the cluster principle she learned the slow way: people did not move, marry, witness, and stand godparent at random. The witnesses and neighbours in one entry are very often the missing relatives in another, and a linked vault lets her see that pattern across entries in a way a single pedigree chart flattens out.
Version history is the quietest feature in the log and, she will tell you, the most honest one. Every time she edits an entry, VaultBook keeps the earlier version, and she can step back through them. Early in her research she was certain — certain enough to write it in the confident present tense — that her German great-grandfather arrived through Baltimore. The version history shows the entry where that certainty lived, the later entry where a passenger list through New York made her hedge it, and the entry where a naturalisation paper naming his arrival date settled it for New York and a different ship entirely. She did not delete the Baltimore claim and pretend she had always known. The trail of her own changing conclusions is still there, dated, in her own words. That matters for two practical reasons. First, if a new record ever reopens the question, she can see exactly what the Baltimore belief rested on — a family story, as it turned out, not a document — and weigh it correctly. Second, it disciplines her now. Writing a conclusion knowing your earlier self will be readable next to it makes you phrase today’s conclusion at exactly the strength the evidence supports, no stronger.
The worked example she returns to when she explains this to beginners at the library’s genealogy evenings is a coal miner she calls, in the log, Thomas Reilly of the patch towns. He is Irish line, second generation, and he moves — or appears to move — in ways that make a tidy timeline impossible until you log the search properly. Her objective, refined three times in the version history, ended as: establish Thomas’s whereabouts in each year from 1895 to 1910, and identify which of the records in that window belong to him and which belong to another man of the same name.
The census gave her two fixed points and one problem. In 1900 he is in a company town in Lackawanna County, a miner, boarding in a household headed by a widow whose surname she recognised from the parish — a cluster link she followed into that widow’s own entry. In 1910 he is eleven miles away, now a head of household, with a wife and three children whose ages back-count to a marriage around 1902. Between those points, the records multiply. A World War I draft card cannot be his — he would have been too old for the registration that card came from, and the log records her checking the registration dates against his estimated birth year rather than just feeling it was wrong. An earlier card, from the registration that did cover his age group, gives a birth date two years off the census estimate, an employer at a named colliery, and a nearest relative whose name she did not know — a thread she labelled and parked under next steps instead of chasing mid-session.
The parish register did the work the census could not. The company town’s church kept baptismal records for the miners’ children, and the log records a search — positive this time — through baptisms from 1902 to 1910 that produced three children, matching the 1910 census ages, with godparents drawn repeatedly from the same two families. Those godparents became entries, then links, then evidence: one godfather appears as a witness on the marriage record she eventually found in the neighbouring parish, the one her place entry reminded her had served the patch town before its own church was built. Without the place entry, she would have kept searching the wrong register. Without the negative-search record, she would not have trusted that the marriage was genuinely absent from the first parish rather than merely unfound. The log entry for Thomas now runs to several thousand words across eight years of intermittent work, and she can hand it to a cousin and say: here is not just what I concluded, but everything I searched, including the searches that found nothing, and here is how my mind changed in 2019 and again in 2023. That is a research log. A conclusion list stapled to a tree is not.
Untangling Two Men With the Same Name
There are, Meg will tell any beginner who will listen, two Thomas Reillys. There are always two Thomas Reillys. In her case there were two men of the same name, born within three years of each other, living in the same Pennsylvania county for two overlapping decades, both miners, both Irish in origin — and, to complete the trap, married to women with the same first name. Online, in dozens of copied trees, they have been merged into one impossible man who marries twice without a death or divorce in between, fathers children in two households in the same year, and is buried in two cemeteries. Untangling them took her the better part of two years. The untangling lives in VaultBook, and deliberately not in her tree, because for most of those two years she did not know the answer, and a tree has no honest way to hold “I do not know yet.”
Her method starts with separation, enforced by structure. Each man gets his own timeline entry, built only from records she can assign to him with a reason she writes down. Not records that mention the name — records that mention the name plus a second identifier that discriminates: an address, an occupation detail, a named associate, an age consistent with one man’s range and outside the other’s. A record that carries the name and nothing discriminating goes into neither timeline. It goes into a third entry, the unassigned pile, which she regards as the most important entry in the whole exercise. The moment you force an ambiguous record onto one man’s timeline because it has to go somewhere, you have started merging them again, this time in your own handwriting.
The timelines are built as plain dated lists, and she builds them in parallel, so gaps show against each other. One Thomas is in the county by 1893, appearing in a city directory at a boarding address, single. The other first surfaces in 1897, already married, in a tax assessment for a different township. For a decade their entries run side by side in her vault — directory listings, census schedules, a mine accident report that names an age and a colliery, church records, children’s baptisms with different godparents, two separate signatures, one man signing with a mark. Read separately, each timeline is thin but internally consistent. Read merged, as the online trees present them, the combined life contradicts itself every three or four years. Laying the two timelines side by side was the first moment the problem stopped feeling like a muddle and started looking like what it was: two ordinary lives that happen to share a name.
On top of the timelines sits the evidence comparison, and she writes it as structured prose and lists rather than leaving it as an impression in her head. For each contested record she sets out, in the same order every time: what the record says, what it does not say, which man it fits, and what would have to be true for it to fit the other one. That last clause does the heavy lifting. A marriage record fits Man A comfortably. For it to belong to Man B, Man B would have to have been in a township where he never otherwise appears, marrying at an age his other records contradict, with witnesses drawn from a family cluster he has no other connection to. Possible is not the test. Genealogy runs on reasonably exhaustive search and the best explanation of the whole body of evidence, and writing out the contortion the alternative requires is how you discover whether your preferred assignment is reasoned or merely first.
The centrepiece is a separate proof-argument entry, and she treats it with a formality the rest of her notes do not get. A proof argument is a written case, addressed to a sceptical reader who is not you: here is the question, here is the evidence, here is why the evidence, weighed properly, supports this conclusion and not the other. Weighing properly means being explicit that records do not all count the same. She walks through the three documents at the heart of her same-name case the way she now teaches others to. A marriage licence application is a record the parties themselves supplied information for, close to the event, but it is still self-reported — ages on licence applications drift, especially when a party had a reason to be older or younger than the truth. A parish register entry was written by the priest or clerk at the time of the ceremony, from direct involvement, which gives it a different and in some respects stronger standing for the fact and date of the marriage itself, though it may say little about ages or birthplaces. A civil death certificate is strong for the fact and date of death — the informant was usually present in the aftermath — and notably weaker for the birth date and parents’ names printed on it, because the informant, often a grieving child, may never have known those things accurately. Same man, three documents, three different levels of trust depending on which fact you are asking each document to prove. Her proof argument says all of this out loud, record by record and fact by fact, before it dares to conclude anything.
Citations are what make the argument checkable, and she cites in VaultBook to a standard she did not bother with in her first five years, when “Ancestry” written in a source field seemed sufficient. It is not, and the reason is not pedantry. A proper citation in her entries names the creator of the record — the parish, the county clerk, the census bureau — then the record type and the specific volume, page, or certificate number, then the repository that holds the original, then the platform through which she accessed it, then the date she accessed it, then the precise locator within the image set, down to image number where a film or browse-only collection requires it. The platform is not the source. A parish register does not become a different record because a website digitised it, and if that platform loses the contract or re-indexes the collection, her citation still points a future researcher — including future Meg — to the actual register in the actual archive. She has already lived the alternative: early notes that said only a platform name and a surname, useless the year the collection moved.
Until the proof argument closes, the unproven line stays exactly where it is: in the vault, labelled, in so many words, hypothesis. She has an entry for the tempting extension of this family — the Irish parish where one of the two Thomases supposedly originated — that carries that label at the top and has carried it for four years. It contains her reasoning, the two records that suggest it, and the record that would settle it and which she has not yet been able to obtain. It is connected by wiki links to both men’s timelines, clearly marked as unproven on every link’s surrounding text. What it has never done is leak into the tree as fact, get exported in a file she shares, or harden by repetition into something she believes because she has seen it written down often. A private vault is the right home for a hypothesis precisely because it asks nothing of it. The tree demands a father for every child. The vault is content to hold a well-labelled maybe, indefinitely, until evidence — not impatience — closes the argument.
Downloaded, Not Just Saved: The Record-Image Habit
Early in her research, Meg lost a record she was sure she had. Not lost the way you lose a paper in a house — lost the way the modern version happens. She had found a death certificate image on a subscription platform, attached a link to it in her notes, built a small conclusion on the informant’s name, and moved on. Two years later she went back to check one detail and the image was no longer there under that link. The collection had been re-indexed, the link resolved somewhere else entirely, and her note, which had carefully recorded the conclusion but not the image, could not prove what she had seen. Nothing about that experience was unusual, and that is exactly why it changed her workflow permanently. A link is not a copy. A link is a set of directions to somebody else’s filing cabinet, written in pencil, in a building you do not control.
The habit she built in VaultBook is blunt in its simplicity: every record image she relies on gets downloaded, renamed, and attached to the vault entry it supports, in the same session in which she finds it. No “I’ll download the important ones later,” because later she will not remember which ones were important, and the platform will not promise to keep them where she left them. The download is part of the research act, the way writing the citation is part of it.
Renaming is where the discipline shows. A platform hands you a file named with an opaque string of numbers, or worse, a generic name identical to the last forty files you downloaded. Left like that, a folder of record images is unsearchable in any human sense — you can find a file only by opening it. Meg renames on arrival to a citation-shaped name: surname, first name, record type, year, and place, in a fixed order, every time. A death certificate becomes a file whose name alone tells her whose it is, what it is, and roughly where it sits in that person’s timeline, before she opens anything. A baptism image, a census page, a draft card — all named to the same pattern, so a decade of downloads sorted alphabetically reads almost like an index. When she attaches the file to the ancestor’s log entry in VaultBook, the name travels with it, and anyone she shares that entry with — a cousin, eventually perhaps a grandchild — inherits the sense of the system without needing her to explain it.
Attaching, rather than merely storing, is the second half of the habit, and it changes what the vault can do for her. A certificate sitting in a downloads folder is a picture. The same certificate attached to the log entry for the man it describes, surrounded by her transcription of its key fields, her note on who the informant was and how much to trust each fact on it, and links to the entries for the people it names, is evidence in context. When she re-reads an entry two years on, the document is there on the page, not in a folder she has to go and hunt through while the thought cools. Scans she makes herself follow the same route: her grandparents’ marriage certificate, a family Bible page photographed at a cousin’s house, a funeral card kept in a drawer since 1968 — scanned, renamed to the pattern, attached to the entries they support, with a note recording where the original physically is and who holds it.
Search reaching inside the scans is what turned this from an archive into a working tool. VaultBook’s search includes OCR, so text inside a scanned document becomes findable rather than merely viewable. Meg’s clearest example is a set of German church records she can read only slowly, with a word list at her elbow. She cannot skim a page of German script the way she skims an English census. But once the images are attached and indexed, searching a surname variant across them surfaces the pages worth her slow attention and lets her skip the rest with a clear conscience. The same applies, more mundanely and more often, to her own typescripts: hundreds of pages of transcribed registers, obituary clippings, and her own past notes become one searchable body. Attachment indexing means the contents of the files she attaches count, not just her entry text — so a name that appears only inside a scanned obituary she attached in 2017 still answers a search in 2026. She has learned to be careful about what OCR cannot do — a badly faded hand, an unusual script, will misread, and she treats an OCR non-hit in a handwritten document as no evidence at all, the digital equivalent of a search she would log as inconclusive rather than negative. But for printed and typewritten material, which is most of her twentieth-century paper, the record collection she built one download at a time is now interrogable as a whole.
The reckoning with her own past arrived through the File/Folder Analyzer. Before the download habit, there were years of scattered saving: record images in a downloads folder, in a folder on the desktop named, optimistically, “Genealogy to sort,” in email attachments from cousins, in a second folder created during one computer change and never merged with the first. Running the analyzer across those locations gave her an inventory of what she actually had — file types, duplicates wearing different names, the same census page saved three times in three places, orphans she could no longer assign to any ancestor at a glance. The cleanup took a winter of Sunday afternoons, working folder by folder: identify, rename to the pattern, attach to the right entry, then delete the stray copies once the vault copy was confirmed. Duplicates were the real revelation. She had been carrying, and backing up, and worrying about, roughly a third more files than she owned in any meaningful sense. What remains is one copy, in the vault, attached to the person it concerns, findable by search and by browsing the entry. The scattered folders are gone, and with them the low-level anxiety that the one image she needed was in whichever folder she had not thought to check.
There is a workflow point underneath the tidiness, and it is the one she presses on beginners. Platforms are access, not custody. Collections move, get re-indexed, go behind different paywalls, lose images in migrations, and occasionally vanish from a site altogether when a contract ends. Every one of those events has happened to at least one collection she uses. None of them, since the habit took hold, has cost her a record she relies on, because reliance, in her system, is defined by what she holds. The image in her vault opens offline, on her machine, whether or not any subscription is current, any site is up, or any company still exists. That is not an argument against the platforms — she uses them constantly, and finds records there she could never reach otherwise. It is an argument about which step of the workflow is the finish line. Finding is not finishing. Downloading, renaming, attaching, and citing is finishing.
The Photo Identification Sessions
A photograph of an unidentified person is not a family treasure. It is a question with a deadline, because the people who can answer it are, by definition, the oldest people in the family, and every year the pool of possible identifiers gets smaller. Meg learned this in the standard way: a shoebox of prints from her mother’s house, perhaps half of them labelled, a quarter labelled with first names only — “Anna and the girls,” which Anna, which girls — and a quarter labelled with nothing at all. She had owned the shoebox for six years before she understood that having the photographs was the easy part. The identification is the perishable asset.
Her sessions in VaultBook’s Photo/Video Explorer start before anyone is asked to remember anything, with the physical object itself. Every print gets scanned twice: front, and back. The back is not a formality. Studio stamps name the photographer and often the town, which dates and places an image within a few years even when the sitter is unknown. Pencil handwriting fades but scans surprisingly well — names, a year, occasionally a relationship, “Mother’s sister,” in a hand she sometimes recognises as her grandmother’s and sometimes does not. Inscriptions on the front margin, embossing, the photographer’s imprint in the corner — all of it goes into the entry. EXIF data plays the same role for the digital end of the collection: the camera files from the last twenty years carry dates and sometimes locations embedded in them, and the Explorer surfaces that information alongside the image, so a cousin’s emailed photo dump stops being a pile of undated files and becomes a rough timeline she can sort and interrogate. For the old prints, the scan itself is the record, and she scans at a resolution that captures the stamp and the pencil, not just the faces.
Each photograph lives in its own entry, and the entry records the identification as an event, not a fact dropped from the sky. Who identified the people in this image. When. In what circumstances. And — the field she insists on, the one most family collections never capture — with what confidence, in the identifier’s own terms. There is a world of difference between her aunt saying “That is your great-grandmother, I would know her anywhere” and the same aunt saying “That might be one of the Kovacs girls, the one who went to Detroit, I think.” Both statements go into the entry, verbatim in spirit, attributed and dated. Five years from now, when the aunt is gone and a cousin challenges an identification, the entry does not just assert a name. It shows the name resting on a certain identification made in 2024 by someone who knew the woman, or on a tentative one made from a faded print at a kitchen table, and future researchers can weigh it accordingly. Labels sort the images by family branch — Irish line, German line, her mother’s people, her father’s — so a session with a relative from one side of the family can be run as a filtered pass through only the photographs that relative has any chance of knowing. Branch labels keep the session inside the witness’s actual knowledge.
The session she describes as the turning point in the whole project happened with her mother’s younger sister, aged ninety-two, over two afternoons with an album Meg had assembled — physically, in print order — from the shoebox’s hardest cases. She worked through the Photo/Video Explorer on a laptop beside the album, opening each image as her aunt reached it, and typed what her aunt said into the entry as she said it. The method matters: she did not interrogate. She asked who, and then she let the answer wander, because the wandering produced the identifications no direct question would have reached. A group portrait outside a company house produced, first, three names, then a story about the house itself, then — because the house reminded her — the name of the man at the edge of the frame whom Meg had not even thought to ask about, who turned out to be the brother who left for Ohio and founded, so to speak, Meg’s own branch. One photograph of a picnic, which three relatives had previously looked at and shrugged over, yielded seven names, a courtship, and the location of a farm that appears in a deed she already held. By the end of the second afternoon the album’s unidentified count had dropped by more than half, and every identification carried its witness, its date, and its confidence, including the honest ones: “I know the face and I cannot give you the name, and it will come to me at three in the morning.” It did, two days later, by telephone, and the entry records that too — identified in session, name supplied by phone two days later, high confidence.
Keeping the identification attached to the image, inside the same entry, is a deliberate refusal of the system she tried first: a spreadsheet. For two years she maintained one — filename in one column, names in another, notes in a third — and it failed in the way separate systems always fail. Files got renamed and the spreadsheet did not. A photograph got shared without its row. Her own filing discipline slipped for a summer and the sheet and the folders quietly diverged until neither could be trusted to describe the other. In the vault there is nothing to diverge. The image, its front and back scans, the EXIF information for digital files, the identification with its witness and confidence, her own notes on dating the image from clothing, cars, and studio marks, and wiki links out to the log entries of every person named in it — all of that is one entry, moving as one thing, searchable as one thing. When she opens an ancestor’s research log, the linked photographs are a click away; when she opens a photograph, the people in it lead back to their logs. The aunt’s voice, in a sense, is filed with the faces she named.
She is unsentimental about the unfinished remainder. A third of the original shoebox is still unidentified, and some of it now will never be identified — the last people who might have known those faces are gone, and the entries say so, recording who was asked, and when, and that they did not know. Even that is worth the entry. It stops the next generation from assuming nobody ever tried, and it leaves the door open the way genealogy always leaves doors open: a face unidentified in Ohio may be named in five years by a cousin in Pennsylvania working from the other direction, on a copy of the same print, in a shoebox of their own.
Interviews That Happen Once
Every other source Meg uses will wait for her. The census will be there next year. The parish register is not going anywhere. An interview with a ninety-year-old is the only item in her entire research programme that happens once, on a day she cannot reschedule, with a source who will not be available for a follow-up session next spring. She treats elder interviews accordingly: as the highest-value and least repeatable work she does, planned like it matters and handled afterwards like evidence, not memorabilia.
Her recording method she calls let-them-wander, and she arrived at it by ruining an early interview with a prepared questionnaire. Twenty tidy questions produced twenty tidy, short, dead answers, and she drove home with less than the same uncle gave her unprompted over coffee a month later. Now she prepares differently: she knows, from the logs, which three or four subjects she most needs from this person — the names of a grandmother’s siblings, life in the patch town before the mine closed, who is in a particular photograph — and she holds them lightly, as destinations the conversation should pass near rather than checkpoints it must hit. She opens with the easiest prompt she owns, usually a photograph or a place, and then she mostly listens. Wandering is not inefficiency. When her uncle drifted from his father’s work to the boarding house his mother ran, the drift itself was the information: the boarders were miners from the same Irish parish the family had left, two of them became godparents, one of them is the reason a whole cluster of her ancestors appears in one county’s records and not another’s. No questionnaire she would have written asks “tell me about the boarders.” The wander found them anyway. Her contributions to a recording are deliberately small — a name repeated back, a “what happened then,” long silences she has trained herself not to fill — because every minute of her voice on the tape is a minute that is not theirs.
The raw recording is never the finished product, and VaultBook’s audio tools are where the finishing happens. Sessions run long — ninety minutes, two hours — and contain what real afternoons contain: a kettle, a doorbell, a long pause while someone searches for a word, a story told twice. She uses the MP3 cutter and joiner to trim the dead air from the head and tail, lift out the unrelated stretches, and split the remainder into segments by topic, joined in an order that follows subjects rather than the afternoon’s accidents. Each segment becomes its own attachment on its own entry, titled by what it is actually about, so that two years later she can listen to eleven minutes on the boarding house without spooling through an hour on other matters to find it. The original full recording stays attached too, untouched, on the interview’s main entry. The segments are the working copy. The original is the record, and she no more edits it out of existence than she would crop a document image to the one line she needed.
Transcription is where her privacy practice stops being a preference and becomes, in her judgement, an obligation. The standard route — upload the recording to a transcription service, get text back — would mean sending a relative’s voice, telling family stories in their own words, to a cloud service’s servers, under terms she has not read and they have certainly never agreed to. She does not do it. VaultBook’s transcription uses Whisper locally, on her own machine, and the recording never leaves it. That distinction is not abstract for her. Her interviewees consented to talk to her, their niece, for the family history. They did not consent to become training-adjacent data on a platform, and several of them, children of an era that taught caution, would have spoken differently — or not at all — if a upload had been part of the arrangement. Local transcription keeps the promise she actually made at the kitchen table: this stays with me.
A machine transcript, she is quick to say, is a draft, not a document. Whisper handles her Midwestern relatives well and struggles exactly where genealogy can least afford errors: names. It renders unfamiliar surnames phonetically, flattens the German names in her other line into plausible English words, and turns mine and parish terminology into near-misses that a careless reader would absorb as fact. So every transcript gets a correction pass, by her, listening at reduced speed with the transcript open in the entry, fixing names against the spellings her logs have established, marking genuinely inaudible passages as inaudible rather than guessing, and noting the two or three places per interview where the speaker contradicts something they said in an earlier session. She does not resolve those contradictions in the transcript. She flags them, links them, and lets the research log carry the argument. The transcript’s job is to report what was said. Deciding what was true is a different job, done elsewhere, on the whole evidence.
Corrected and searchable, the transcripts join the rest of the vault’s web. A transcript entry is linked to the log entry of every ancestor discussed in it, so opening her great-grandfather’s log shows not just census and register evidence but a link to his grandson, on tape, describing the man’s habits, his temper, the songs at his funeral — material no record office holds. Search across transcripts has twice produced the answer to a question she was working in a completely different entry: a place name mentioned in passing in an interview she had filed under another branch entirely, surfacing because the words were text now and the vault searches its text as one body. This is the compounding return on the unglamorous correction pass. An uncorrected, unlinked audio file is a keepsake. A corrected, searchable, linked transcript is a source, citable like any other — she cites interviews in her logs with speaker, relationship, date, place, and the segment and timestamp for the passage relied on — and usable by researchers who will never hear the recording, including, one day, descendants for whom these voices belong to people five generations back.
Consent and sensitivity govern the last step, and she handles them at the level of the individual passage, not the whole interview. Most of what her elders tell her is freely shareable within the family, and she says so when she asks permission, and asks. But a long, wandering interview with someone who trusts you does not stay on safe ground. It reaches the relative who is estranged and why, the child given up, the marriage that began earlier than the certificate arithmetic politely suggests, the feud, the bottle, the money. Some of that concerns living people. Some of it would wound people she loves if it travelled as a casual family anecdote with her name lending it authority. Her rule is that such passages do not go into the shared, plain-text body of her work at all. They stay in entries — or clearly marked sections — held under VaultBook’s per-entry encryption, locked with her passphrase, in the same vault as everything else but readable only when she deliberately unlocks them. She can still find them, still link to the entry, still know exactly what she holds and why. What she cannot do is accidentally include one in an exported file, a shared printout, or a moment of careless generosity at a family gathering. Encryption here is not about hackers. It is about making the sensitive material require a deliberate act to open, every single time, so that sharing it is always a decision and never a slip.
She keeps a short note in each interview entry recording the terms on which it was given: that the speaker agreed to be recorded, what they understood the recording was for, and any limits they set — and her elders do set them, once they trust the recorder will still be there in an hour. One aunt gave her a story with the words “not while your cousins are living” attached, and that restriction sits at the top of the segment’s entry, in plain sight, where any future version of Meg will meet it before meeting the story. People speak to the future in these interviews. The least the future can do — and the vault, organised this way, actually does it — is remember the conditions under which they spoke.
The DNA Annex: Notes That Are Not Ready to Be Family Knowledge
Meg Ellison keeps her DNA work in a separate corner of the vault, on purpose. The research logs, proof arguments, record downloads and photo identifications described earlier are private in the ordinary sense: unfinished, sometimes wrong, not meant for cousins to browse while she is still working. The DNA notes are private in a different sense. Some of them, if they turned out to be right, would change what living people believe about who their parents were. That is a different category of knowledge, and Meg treats it that way in the structure of the vault itself, not just in her intentions.
She calls the section the DNA Annex. It holds four kinds of material, and only the fourth is encrypted entry by entry. The first three live as ordinary vault entries, because they need to be easy to open and revise during a working session. The fourth — five entries — sits behind its own passphrase protection, each encrypted individually with per-entry AES-256-GCM encryption. The distinction is the whole design: she did not want a vault that demanded a ceremony every time she checked a census transcription, and she did not want her most consequential suspicion sitting in the same plain layer as a note about which cousin identified the woman in the church picnic photograph. Per-entry encryption lets both requirements be true at once.
The first kind of material is match cluster work. Meg long ago stopped treating her match list as a list; she treats it as a population to be sorted. Her cluster notes group matches by shared segments and shared matches into provisional families: a cluster that clearly belongs to her mother’s Ohio line, one that belongs to her father’s people in Pennsylvania, a stubborn cluster in West Virginia that fits neither side cleanly and has sat unresolved for two years. Each cluster gets an entry recording which matches anchor it, what their public trees show, where the trees agree and contradict each other, and what the next test would be — which record, which additional tester, which courthouse — that could move it from provisional to placed.
This work is methodical and, on its own, not sensitive. It is also work the testing platform does badly as a long-term home: it shows her matches, shared matches and trees, but gives her no durable place to write her own reasoning in her own structure, revise it over years, and keep it when her attention or her membership lapses. Her cluster entries are that place. A cluster is a hypothesis about grouping, not a conclusion about parentage, and Meg writes cluster notes in that register — observed, grouped, unplaced, next step — the same disciplined register she uses in her research logs.
The second kind of material is shared-match analysis, the working layer beneath the clusters. For a match she cannot yet place, she works through who else matches both of them, what surnames recur in those shared matches’ trees, and whether the group points toward a known ancestral couple or toward a gap in her own tree. These entries are dense, provisional and frequently revised, with dated additions kept rather than overwritten, because in DNA work the history of how a conclusion formed matters almost as much as the conclusion. A match that looked like noise in 2023 became the key to a cluster in 2025 when two more testers appeared in the same group; tidying away the early entry would have lost the thread of her own reasoning.
The third kind is the correspondence log, and anyone who has done DNA genealogy will recognise its frustrations immediately — they are among the most discussed irritations in the field. Some matches reply within a day, generously, with family Bibles and photographs. Some reply after four months with two sentences. A large share never reply at all. Her log records, for each outreach: the date, the match, what she asked, what identifying detail she offered about her own line to establish good faith, and the outcome. The point is not grievance; it is memory. With several hundred matches worth contacting over the years, she cannot rely on recollection to tell her whether a silent match is someone she has already approached twice or never written to. The log prevents duplicate pestering, which matters ethically as well as practically: a stranger who did not answer the first message has given her an answer of a kind, and she does not send a third.
The log also records a structural frustration of platform messaging every active user learns the hard way: the lapsed member problem. A match whose paid membership has expired can typically see that a message exists — the notification, the envelope, the prompt to rejoin — but cannot read it until they resubscribe. From Meg’s side, the message shows as sent and unread, indefinitely; she cannot tell whether she is being ignored, whether the person tested once on a holiday whim and never returned, or whether they are staring at a locked inbox. Her log has a status category for exactly this: sent, possibly unreadable. It stops her drawing conclusions about people’s willingness from what may simply be a paywall between a message and its reader, and it shapes her strategy: for important matches she notes whether an alternative route exists — a public tree elsewhere, a surname society, an obituary naming survivors — and why she did or did not pursue it. Sometimes the honest entry is: no other route found; will check again in six months. DNA genealogy runs on timescales of years, and the annex is built for that pace.
Then there is the fourth kind, the sensitive core, and it is small: five entries. The encrypted core is not where DNA work happens; it is where one particular kind of finding waits while she decides what it is and what, if anything, anyone should do about it.
The finding, stated as carefully as she states it to herself, is a suspected not-parent-expected event — the term genealogists use when DNA evidence suggests that a person’s documented parent is not their biological parent. The suspicion sits two generations back and points sideways into a branch with living descendants who believe, as she once believed, a straightforward story about who their grandfather was. The cluster evidence does not fit the documented story. It fits another, involving a man who lived in the same township, appears in the same church records, and has never been connected to the family in any paper source she has found. The evidence is suggestive, not conclusive. It may never become conclusive, because that would require either a document that probably does not exist or a test from a specific living person who has not tested and may never agree to.
Those five entries hold, respectively: the cluster analysis in its full unsanitised form, including match details she omits from her ordinary notes; the timeline comparing the documented story against the DNA-implied story, date by date and place by place; the alternative explanations she has considered and what evidence weakened or survived each; the names of the living people who would be affected if the suspicion is correct, and her private assessment of what each knows, suspects or has said; and the protocol entry, the most important of the five.
The protocol is a document Meg wrote before she needed it, in a calm hour, so she would not have to improvise her ethics in an emotional one. It states what evidence would count as confirmation: not a stronger feeling, not one more circumstantial match, but a defined threshold — a match at a specified closeness to a specified line, or a document of a specified kind, or both. It states what would count as refutation, because a protocol that can only ever confirm is not a protocol, it is a countdown. It states who would be told, in what order, if the threshold were met: first the one living person whose own parentage is directly implicated, to hear it from family rather than discover it from a database; then, only with that person’s agreement, the next circle. It states the manner: in person or by voice, never by forwarded screenshot, never in a group message, never as an aside at a reunion. And it states what never gets written into any shared tree, threshold met or not: the suspicion itself, in any form, attached to any profile, because a tree note propagates in ways its author cannot recall — copied into other trees, indexed, screenshotted, repeated by people who never saw the caveats. Some things, Meg’s protocol says in its first line, are told to people. They are never published to a database.
Why encrypt these five entries individually rather than locking the whole vault? Meg’s answer is practical, and it generalises. The ordinary work is in constant use — at her desk, at the library on her laptop, at a courthouse on her phone. Friction applied to all of it would not make the sensitive material safer; it would make the whole system slower, and slow systems get circumvented: notes migrate to sticky notes, to email drafts, to whatever is open. Per-entry encryption inverts the burden. The ninety-five percent of the vault that is merely private opens with the ease of a local file, because it is a local file, under her custody, never uploaded anywhere; the five entries that are genuinely dangerous to the wrong reader carry their own AES-256-GCM encryption, each sealed on its own, requiring a deliberate act to open. She opens them rarely — at home, alone, when she is specifically working on that question. The rarity is itself part of the protection: they are not casually on screen, not one stray click away during a session about something else, not visible over her shoulder at the library.
Per-entry encryption also matches the moral structure of the material. The cluster notes concern her own research process; the five core entries concern other people’s identities — people who did not test, did not consent to be analysed, and in two cases do not know the question exists. Every time she types the separate passphrase, she is reminded of what kind of material she is about to handle. Meg is sceptical of tools that promise to make ethical decisions for their users, but not of tools whose structure makes the weight of a decision felt at the moment it is taken. The extra step is a threshold, in both senses.
That point about non-testers is the ethical centre of DNA genealogy. Every match implicates people who never spat in a tube: a match between Meg and a stranger is also evidence about the stranger’s living parents, siblings and children, none of whom agreed to feature in her research, and the man in the alternative story has descendants who did not ask to be written into anyone’s family secret. Her rules reflect this. Living people’s full names appear in the encrypted core only where the analysis genuinely requires them; ordinary entries use initials and match usernames. She does not record living matches’ locations more precisely than the platform displays, and she does not copy living people’s photographs from social media into the vault, however useful they would be, because usefulness is not consent. Her protocol’s order of telling rests on one principle: the people whose own story it is hear it first, from a person, with the evidence and its limits laid out honestly — including the limit that DNA suggests and documents decide, and that she may simply be wrong.
She may be wrong. The annex is designed to hold that possibility open for years if necessary. That is what it is for: not secrecy as drama, but patience as method. A suspected not-parent-expected finding is not a discovery to announce; it is a question to carry carefully, in a place that is fully hers, encrypted at exactly the points where carelessness would do harm, until the evidence resolves it — or until it passes, still unresolved and still unwritten in any shared tree, out of her hands entirely. The vault does not make that judgment for her. It gives the judgment a safe place to wait.
Why VaultBook Beats the Market-Leading Note Apps for This Work
Meg did not arrive at VaultBook by disliking the other tools. Over fifteen years she has kept genealogy notes in nearly every major note app, and her working papers include exports from most of them. Each migration was prompted by the same discovery, made with a different tool each time: the app was excellent at notes in general and wrong, in one specific mechanical way, for notes like hers — notes full of record images and audio, notes about living people, notes that must survive a lapsed membership and a dead laptop, notes containing one suspicion she would not want read aloud. This section makes that argument app by app, against the market-leading note apps a genealogist would realistically choose instead: Notion, Evernote, Microsoft OneNote, Google Keep and Apple Notes, with Ancestry at the end as the custody comparison that matters most in this field. For each, the claim is narrow and mechanical — why VaultBook is better for this scenario, this reader, this work — and the honest counterweight is stated too, because every one of these products is genuinely stronger than VaultBook at something real.
The mechanisms VaultBook argues from are the same five throughout, applied differently each time. Custody: the vault is a file on Meg’s devices, never uploaded, because there is no account holding her data and no server her notes pass through. Per-entry AES-256-GCM encryption: individual entries can be sealed with their own protection while the rest of the vault stays friction-free, as the DNA Annex depends on. Offline reliability: the vault opens with zero network requests, in a courthouse basement or a cemetery, exactly as at her desk. One-file portability: the whole working life of the research — logs, images, audio, PDFs, proof arguments — travels as a single file she can copy to any device with a browser or hand to her executor on a drive. And independence: no account in good standing, no sync service, no company’s continued existence stands between Meg and her own notes. Watch how those five mechanisms land differently against each rival, because the differences are the argument.
Notion. Notion is the tool Meg’s younger fellow researchers most often recommend to her, and she understands why; she kept a research workspace in it for over a year. The mechanical problem is custody and the shape of its encryption. Notion encrypts data in transit with TLS 1.2 or better and at rest with AES-256, but the keys are managed by Notion, and the product is not end-to-end encrypted. That distinction is not pedantry. It means the protection guards against outsiders intercepting or stealing storage, while Notion itself retains the technical ability to read workspace content — and its own policies acknowledge a version of this: support personnel can access workspace data with the user’s consent for troubleshooting. The safeguard is a policy control, a promise about when access will happen, not a cryptographic impossibility. For a shopping list that is a perfectly reasonable design. For the DNA Annex’s five core entries — a suspected not-parent-expected finding naming living people — Meg is being asked to rest the most sensitive sentences she has ever written on a company’s internal access policies and their continued good administration. VaultBook’s per-entry AES-256-GCM encryption rests them on mathematics she controls: an entry sealed in her file, on her device, with a key derived from her passphrase, is not readable by VaultBook’s support team with her consent or anyone else’s, because there is no server copy for anyone to consent over. That is the concrete advantage for this scenario, and it is not close.
The second Notion problem is offline, and it is the one that actually ended Meg’s experiment, on a research trip rather than in a privacy debate. Notion’s offline support works through per-page downloads in its desktop and mobile apps: you download the pages you expect to need. Subpages do not automatically come with a downloaded parent page. A downloaded database shows only the first fifty rows of its first view. Embeds, AI features and forms need a connection. Now picture the genealogist’s field conditions. Meg’s research log is long — years of entries, far past fifty rows in the view she works from — and her method depends on following links from a log entry to a proof argument to a record image: subpages and embeds, precisely what offline mode does not guarantee. She prepared for a courthouse trip by downloading what she thought she needed, arrived in a basement records room with no signal, and found the entry she needed referenced three subpages she had not individually downloaded, in a database view truncated above the row she wanted. The notes existed; they were simply not there. VaultBook’s offline reliability is not a mode to prepare in advance; it is the product’s only state. The file is on the device, opens in the browser with zero network requests, and every entry and attachment inside it is present because there is nowhere else for any of it to be. For work in archives, cemeteries, small-town libraries and elderly relatives’ homes, that is not a feature comparison. It is the difference between the notes travelling and the notes staying home.
Honesty requires the counterweight, and it is substantial: Notion is genuinely stronger than VaultBook at live collaboration and shared workspaces, and at beautiful publishing. A family history society building a shared, multi-editor research hub, or a genealogist publishing an elegant public page about a line, will do it better in Notion, and VaultBook — which has no live collaboration at all — should not pretend otherwise. Meg’s answer is that collaboration is exactly what her working vault must not be: the shared, polished layer of her genealogy lives elsewhere, downstream, after conclusions settle. The layer where suspicions are tested and living people are discussed goes in the tool whose maker cannot read it and whose offline mode is not a partial download. Both judgments can be true because they concern different layers of the work.
Evernote. Evernote is where Meg’s digital genealogy notes began, and its failure mode for this work is the granularity of its encryption. Evernote’s encryption feature protects selected text inside a note, on desktop, using AES-128. It does not encrypt the whole note, and it does not encrypt attachments. Hold that against the genealogist’s actual material. The sensitive unit in Meg’s work is almost never a paragraph. It is the attachment: the record image, the interview audio, the PDF of a fragile letter, the photograph with names pencilled on the back. A tool that lets her encrypt a sentence while the deed scan beside it sits unencrypted has protected the caption and exposed the evidence. It also cannot protect the structure of her most sensitive work: the DNA protocol entry is sensitive as a whole — its headings, its order of telling, its list of who would be affected — not as one selectable passage inside an otherwise safe note. VaultBook’s per-entry encryption seals the entry entire, attachments and text together, with AES-256-GCM. The five core entries of the annex are protected as units because they are units.
Evernote’s current free tier sharpens a second, practical point about custody, though the argument here is about limits, not price models. The official plan comparison caps the free plan at fifty notes, one notebook and one device. Meg’s vault holds thousands of entries across decades of work, consulted from a desktop, a laptop and a phone; in a system where the archive’s size and device count are set by plan tier, access to her own accumulated research is a function of an account’s standing. VaultBook’s one-file portability answers this mechanically: the file is the archive, it opens on any device with a browser, and copying it to a third device is a file copy, not a licensing event. The honest counterweight is real: Evernote’s web clipper is mature and excellent, the best Meg has used, and its long cross-platform history means old notes have generally survived the years. A genealogist whose primary activity is clipping web material will feel that strength daily. But clipping is intake. The vault is where intake becomes a lifetime archive with a sensitive core, and for the archive, whole-entry encryption and unmetered custody of her own file are the mechanisms that decide.
Microsoft OneNote. OneNote’s protection model is the closest of the cloud notebooks to what Meg needs, which is exactly why its limits are instructive. Password protection in OneNote applies at the level of a section, using 128-bit AES encryption. A section is a blunt unit: larger than the entry, organised by the notebook’s structure rather than by sensitivity. Meg’s sensitivity does not follow her notebook structure. Her DNA work sits in one area of the vault, but within it, ordinary cluster notes and the five core entries are interleaved by topic — the core entries are the deep layer of the same subjects the ordinary entries introduce. Section-level protection forces a choice she does not want: either quarantine the sensitive material into a separate section, advertising its existence and severing it from its context, or protect an entire section and accept friction on all the ordinary work inside it. Per-entry encryption in VaultBook follows the sensitivity, entry by entry, wherever the entries naturally live. Two further OneNote specifics matter in the field. Protected sections are excluded from search until they are unlocked, which means the protected material cannot be found by the vault-wide searching a researcher does constantly — Meg searches her whole vault for a surname before every courthouse trip, and material that vanishes from search until unlocked is material she will, in practice, fail to consult. And recordings cannot be password-protected at all. Meg’s locally transcribed interviews exist in the vault as audio alongside their transcripts; in OneNote the audio of an elderly relative describing family matters could not sit inside the protection at all, however the surrounding text was arranged.
OneNote’s storage model completes the picture. Notebooks in current versions live in the cloud through a Microsoft account, on OneDrive or SharePoint, with offline edits syncing when a connection returns. For most users that is convenience. For Meg it means the canonical home of her notes is a server, her access runs through an account, and the offline copy is a cache of a cloud original rather than the thing itself. VaultBook inverts it: the file on her device is the original, there is no account and no server, and nothing syncs because there is nothing to sync to — copies are copies she makes deliberately, to places she chooses, which for material about living people is a feature and not an inconvenience. The honest counterweight: OneNote’s free-form canvas and handwriting and ink support are genuinely superior, and its integration with Microsoft 365 is deep and real. A genealogist who sketches pedigree layouts by hand on a tablet, annotates record images with a pen, and lives inside Word and Outlook will get things from OneNote that VaultBook simply does not offer. Meg annotates in her tree program and writes in text; for her workflow the canvas is surplus and the section-level lock is a daily constraint. The comparison turns, as it should, on which mechanisms the actual work touches.
Google Keep. Keep is the tool on Meg’s phone’s home screen — for other things. Its position here is the quick-capture layer, and the scenario shows why that layer must not become the archive. Google Keep offers no per-note password, lock or per-note encryption; its own help documents no such feature. Data is encrypted in transit and at rest within Google’s data centres, which protects it from outsiders, while Google itself processes content — its systems handle handwriting recognition and search across your notes, so the content is machine-readable by the provider as part of how the product works. Export runs through Google Takeout, delivering your data as an export event rather than as a living, working file. Now run the scenario. A cousin, mid-conversation at a reunion, mentions that her mother always said the family name was spelled differently before the war, and that there was a brother nobody listed. That fragment goes into quick capture — and an hour later turns out to be the first thread of the annex’s central question. In Keep it sits in a system with no way to seal it, processed for handwriting and search, in an account-based store whose export is a snapshot. In Meg’s practice the fragment is transcribed the same evening into the vault, where, if it belongs to the sensitive core, it becomes part of an entry sealed with per-entry AES-256-GCM encryption, in a file no provider processes because no provider ever receives it. Custody is the whole mechanism: a note never uploaded cannot be processed, indexed or read by anyone’s systems, including helpful ones.
The honest counterweight is easy, because Meg uses it herself: Google Keep is the fastest quick capture on Android, and sharing a list from it is effortless. For a grocery list or a shared packing list, nothing here argues against Keep, and VaultBook — a considered working environment rather than a two-tap capture surface — is worse at that job. The argument is only about where fragments about real people’s parentage come to rest. Quick capture is a doorway; the vault is the room. Genealogy goes wrong when the doorway is mistaken for the room, and a tool with no lock on any note should not be the room where the DNA Annex lives.
Apple Notes. Apple Notes requires the most careful treatment in this section, because it contains the one rival protection that deserves an unqualified concession. Locked Notes in Apple Notes are end-to-end encrypted, using AES in Galois/Counter Mode with a key derived through PBKDF2 from the user’s password. That is real end-to-end encryption, and this article concedes it plainly: for a locked note consisting of plain text, on Apple devices, Apple Notes offers cryptographic privacy of the same class VaultBook claims. Any comparison that waved this away would be dishonest, and genealogists on Apple hardware are entitled to hear it stated straight.
Then the mechanism that decides the genealogy case, and it is not the cipher. It is the attachment exclusion. Apple Notes cannot lock notes that contain video, audio, PDF or file attachments. It cannot lock shared notes, and it cannot lock notes synced in from other accounts. Consider what a genealogist’s vault actually is. It is attachments. The record images are the evidence. The interview audio is the source. The PDFs are the fragile originals. The photographs are the identifications. Meg’s most sensitive entries are not paragraphs of prose that happen to mention evidence; they are bundles in which the attachment is the point — the timeline entry is inseparable from the record images it compares, the interview-derived entry from its audio. In Apple Notes, every one of those entries is, by rule, unlockable: the moment the evidence is attached, the end-to-end protection becomes unavailable for that note. The protection covers precisely the notes a genealogist needs least protected — plain text — and excludes precisely the notes she needs most. VaultBook’s per-entry AES-256-GCM encryption has no such exclusion in Meg’s use: the entry seals as a whole, text and attachments together, because the encryption is a property of the entry in her file, not a feature layered over a sync service with content-type exceptions. That single mechanical fact — attachments excluded there, attachments included here — is worth more to this scenario than any other comparison in this section.
The platform limits compound it. Apple Notes exists on iPhone, iPad, Mac and Vision Pro, and on the web through iCloud; there is no Windows or Android application. Meg’s research life crosses platforms constantly — a Windows machine at one archive, an Android phone in her pocket for two years of the project, a cousin’s computer, a library terminal. A vault that can only be opened inside one company’s hardware ecosystem is a vault whose availability depends on her next purchasing decision. Notes locked with a device passcode add a further wrinkle: they cannot be viewed on iCloud.com, so the one non-device route into the notes is closed to exactly the locked material. VaultBook’s one-file portability is the direct answer: the file opens on any device with a browser, because it was never an application-specific database in a single vendor’s cloud; it is her file, and browsers are everywhere. The honest counterweight, beyond the encryption concession itself: on Apple devices, Apple Notes is effortless — instantly available, beautifully integrated, and its locked plain-text notes are genuinely private. A genealogist who works entirely within Apple’s ecosystem and whose sensitive material is truly text-only could reasonably choose it. Meg’s material is not text-only, and her devices are not all Apple. The exclusion decides.
Ancestry, and the custody comparison. Ancestry is not a note app, and it is included because for genealogists it is the comparison that dwarfs the others: the platform where the tree lives and where the records come from. Meg’s tree program is her system of record for conclusions, and Ancestry is her largest source of record images. Ancestry’s own help documentation states the custody terms plainly, and every genealogist should read them before they need them. If you cancel a paid membership, you keep a free guest account; your trees remain, and remain editable. But you lose the ability to view most records — including record images you previously found and attached to people in your tree. The images stay linked to the tree; they are simply not visible to you until you resubscribe. Records you downloaded yourself, to your own storage, remain yours. Read that again with a researcher’s eye. Years of work — the census page attached to the grandfather, the draft card attached to the uncle, the parish image attached to the marriage — can remain structurally present in your tree and practically invisible to you, because what you attached was never the record. It was a link to the record, and the link’s visibility is a function of an account’s standing.
This is the scenario VaultBook’s custody mechanism answers most directly, and it is why the record-download workflow matters as much as it does in Meg’s practice. When she finds a record that matters, she downloads the image itself, and that image goes into her vault — into the single file, on her devices, under her control, in several entries’ cases under per-entry encryption. That image cannot go dark. No account standing or platform decision stands between Meg and a file she holds; the vault makes zero network requests and answers to no server, so the evidence she gathered at sixty-one remains openable at eighty-one on whatever machine she then owns, and it passes to her executor as a file, not as a login someone must think to maintain. The honest counterweight is that Ancestry’s record database, its hint system and its sheer scale are genuinely unmatched — VaultBook has no records to discover, and the next section says so without hedging. Discovery is the platform’s job, and it is magnificent at it. But discovery and custody are different jobs. The platform finds the record; the vault keeps it. A genealogist who lets the platform do both has, on the platform’s own stated terms, an archive that can be switched to invisible while remaining technically intact. Meg’s rule follows from the mechanism: everything the tree links to, the vault holds. Links break, lapse and darken. Files she holds do not.
Across all six comparisons the pattern is the same. Each rival is a fine product whose centre of gravity is somewhere else: collaboration, clipping, canvas, capture, ecosystem, discovery. The genealogist’s working vault has a specific centre of gravity — attachments as evidence, living people as subjects, decades as the timescale, archives and cemeteries as the workplace, and a small core of material that must be protected as whole entries rather than as sentences, sections or text-only notes. On custody, on per-entry AES-256-GCM encryption that includes attachments, on offline reliability with zero network requests, on one-file portability to any device with a browser, and on independence from any account or server, VaultBook is not marginally better for that centre of gravity. It is the only tool in this list built on it.
The Honest Limits of the Vault for Genealogy
Everything claimed above needs its boundary stated in the same article, so here it is without hedging. VaultBook does not draw pedigree charts. It does not lay out a family in generations, render descendant reports, or produce the visual tree that Meg shows her cousins at reunions. It does not compute relationships: it will not tell her that a match is a probable second cousin once removed, calculate a shared ancestor from two lines, or check whether a proposed relationship is consistent with the tree she has built. Those are computations over structured family data, and the vault holds working notes, not a structured tree.
It has no hint engine and no record database. Nothing in the vault will nudge her toward an unexamined census collection, suggest a possible parent from a name and a county, or surface a record she did not already know to look for. Discovery — the part of genealogy that feels like treasure hunting, and the part the platforms do superbly — does not happen in the vault at all. Every record image inside it arrived because Meg found the record elsewhere, judged it worth keeping, and downloaded it herself.
There is no live collaboration with cousins, and no automatic sync with her tree. A cousin cannot comment inside an entry, co-edit a proof argument in real time, or watch a conclusion update as she types. When a proof argument settles and a conclusion changes, Meg carries that conclusion to her tree program herself, by hand, and cites the vault entry it came from. The vault and the tree can drift if she is careless, and the discipline of reconciling them is hers, not the software’s. For some researchers that manual step will feel like a defect. Meg has come to treat it as the checkpoint where a working suspicion becomes a stated conclusion — a deliberate crossing, made by a person, rather than a silent propagation. But it is a real cost in time, and this article will not pretend otherwise.
Nor does the vault replace the platforms’ social machinery: no messaging with matches, no shared trees to browse, no community forums, no DNA tools of its own — no segment data, no chromosome browser, no ethnicity estimate. The DNA Annex organises her thinking about results produced elsewhere; it produces no results.
Two further limits deserve naming because researchers feel them in the first month. The vault does not transcribe for her: the handwriting in a parish register becomes searchable text only when Meg types it, and the interview audio becomes a transcript only through the patient local work described earlier in this article. And it does not back itself up by magic — one-file portability makes copying simple, but the copies exist only because she makes them, on her schedule, to drives she controls. A researcher who never copies the file has a single point of failure, however private that file is. Privacy and safety are related, but they are not the same discipline, and the vault automates only one of them.
That is the honest shape of it. The tree program keeps the conclusions and draws the charts. The platforms hold the records and find the matches. The vault is the private research layer beside them — the logs, the evidence, the reasoning and the five sealed entries — doing one job neither of the others is built to do, and none of theirs.
Building the Vault in a Weekend
Meg did not build her vault in an afternoon of enthusiasm and then spend a month fixing it. She built it in a weekend, in order, with the tree program closed for most of Saturday morning. A vault that starts with structure, then the most load-bearing records, then a clear rule for what gets encrypted, becomes a place she can trust on a tired Tuesday night.
Friday evening was for decisions, not importing. She opened VaultBook — a single HTML file on her laptop, running locally, making no network requests — and resisted dragging folders into it. Instead she wrote down what the vault was for and what it was not for. It was for the working layer: research logs, proof arguments in progress, photo identifications, interview notes, DNA match notes, copies of records she relied on, and unanswered questions. It was not a replacement for her tree program, which remained the system of record for the cleaned, sourced tree of roughly 2,400 people, and not a place to store a second copy of every hint image she had ever saved.
The label architecture came next, still on Friday, because labels are cheap before there is content and expensive after six hundred entries depend on them.
Surname labels: one for each direct-line surname she actively researched — eleven on day one — not for every surname that had ever married into the tree. Collateral surnames earned a label only when she opened a real research question about them, keeping the list from becoming a second index of the tree.
Place labels, at county level, because county is where American records live: the Ohio county where her father’s people farmed, the Pennsylvania county her McCreadys left in the 1880s, the county whose courthouse she visited most often. Town labels only where a town had its own archive she returned to. Place labels answer: “If I am going to this repository, what else do I need while I am there?”
Record-type labels: census, deed, probate, vital record, church record, newspaper, photograph, interview, DNA, correspondence, and map. No separate labels for birth versus death certificates — the entry title carried that detail and search could find it. Labels were for filtering a working list down to something actionable, not for cataloguing for its own sake.
Status labels, the ones she would use most: open question, waiting on record, in proof, ready for tree, sensitive — do not share, and follow up. Every research log was expected to carry a status label at all times, and changing it as work moved was part of closing a session.
She tested the architecture on twenty entries she knew she would create — a census image, an unidentified shoebox photograph, a DNA match note, a badly transcribed deed. None needed a fifth family. Search would do the rest: VaultBook searches titles, body text, OCR text from images, and indexed attachment content, so a name buried in a transcribed letter needed no label to be found. Labels were for the views she returned to — Kanban board, filtered lists, sidebar.
Saturday morning was templates. Meg created four template entries, labelled template, to copy each time she started new work, rather than retyping a structure from memory and letting it drift.
The research log template came first. At the top: the question as a single answerable sentence — not “find out about Grandpa Hart” but “Was the John Hart who bought land in the county in 1874 the same John Hart who married Mary Ellison in 1876?” Then the date opened and the people concerned, with tree identifiers. Then a dated work log: what she searched, where, with what variants and ranges, and — critically — what she did not find, because recorded negative searches stop her re-running the same census search eighteen months later with the same wrong birth year. Then sources, cited in full as they should appear in the tree. Then findings, kept separate from the work log, because activity and conclusions are different things and confusing them is how a hunch becomes a fact. At the bottom: next steps, at most three, each concrete — “order the probate packet,” not “look into probate.”
The proof argument template was stricter, because it is where Meg convinced herself, in writing, before changing the tree. It opened with the question and the conclusion she believed the evidence supported, in two or three sentences. Then one section per record: citation, what the record actually said, how reliable it was for this fact, and what it did not establish — a census proves residence and approximate age, not parentage, and the template forced that sentence out each time. Then a conflicts section that could not be left blank: any disagreement on a birth year, birthplace, or spelling was named and resolved with reasons, or the argument was marked not ready. Then the conclusion, and a final line: what would change my mind.
The photo ID record template served the shoebox work. At the top: the physical photograph described — size, card stock, photographer’s mark, and where the original lived, and whether Meg held the original or only a scan. Then the image, attached. Then identifications as a dated list: who, identified by whom, on what date, with what confidence — “Aunt Ruth, 14 October, identified the woman at left as her mother, certain, recognised the brooch.” An identification without a source is a guess with good manners. Then dating evidence — clothing, a car, a known child’s age, a studio’s years — and related notes linked to the logs and interviews mentioning the same people.
The match-note template governed DNA work, one entry per match or cluster. At the top, the match in Meg’s own words — shared amount, predicted relationship, side of family if known — because she needed her own analysis, not a wholesale copy of another person’s private data. Then the paper trail, with citations to records, not other people’s trees. Then contact history: dates written and replied, what was agreed about sharing. Then hypotheses, labelled as hypotheses, and what record or test would confirm or refute each. Finally a sensitivity line: who could be affected if this note were wrong, or shared prematurely. That line changed how she wrote everything above it.
Saturday afternoon was the audit, and she imported nothing until it was done. She ran the File/Folder Analyzer over a decade of scattered material: a downloads folder become a second archive, desktop folders named “genealogy,” “genealogy 2,” and “to sort,” an external drive from a 2019 courthouse trip, email attachments saved in three places. The Analyzer showed file counts, types, sizes, and where duplicates clustered. She decided in advance what belonged in the vault, what belonged only in the tree program, and what she could delete because the record lived, properly cited, in both. Audited, a decade of accumulation is mostly duplicates and almost-finished transcriptions — and knowing that stopped the vault inheriting the mess.
Batch organisation followed, folder by folder, never all at once. For each folder: create or update the entries the material belonged to, attach each file to the entry it supported — a deed image to the log that cited it, a letter scan to its correspondence entry with its transcription — and only then clear the loose file. A scan named IMG_2041 means nothing in eighteen months; the same scan attached to a photo ID record, with Aunt Ruth’s identification dated and sourced, means something forever. Where a batch formed a set, she used the vault’s PDF tools to merge a probate packet into one ordered document, split an over-long combined scan back into separate records, and compress the few enormous scans. Each thing arrived with its reason for being there already written down.
Sunday was for the online tree’s most load-bearing images, ruthlessly prioritised. She downloaded first the images attached to direct-line ancestors: birth, marriage, and death records for parents, grandparents, great-grandparents, and on back, plus the census, deed, and probate records she had actually relied on in a proof argument. If the service disappeared, her access failed, or a record set was withdrawn, the evidence under her direct line would still be in her hands, attached to the entries that explained it. Unverified hints and distant collateral images stayed on a list for a slower pass. Checking each image against its citation as she filed it, she found two citations pointing at the wrong image and one record cited for years from a transcription; the original disagreed on a middle name, and both vault and tree were corrected, with a short proof entry explaining the change.
The GEDCOM question settled itself once Friday’s boundary was clear. She exported a fresh GEDCOM on Sunday, stored it inside the vault as a document entry labelled with its export date, and kept the tree program’s own native file where it had always lived, on its own backup schedule. The entry noted when it was exported, from what version of the tree, and what had changed since the last export. She would refresh it at milestones, not weekly: the vault did not need to mirror the tree, only to show her, offline, what the tree claimed on the day a question was opened.
Last, Sunday evening, encryption — entry by entry, under one rule: encrypt anything whose disclosure would harm or embarrass a living person, or expose an unproven conclusion about people living or recently living. On day one that meant all DNA match notes, encrypted per entry, AES-256-GCM applied individually rather than to the whole vault. It meant notes on living people: the cousin whose parentage question was open, interview passages given in confidence, contact details. It meant one proof argument still in progress, whose conclusion would land in a family group chat as settled fact. Those she encrypted, tested she could reopen them, and listed in a plain index entry which entries were encrypted and why.
Everything else stayed plain, deliberately: census transcriptions, deed abstracts, published obituaries of people long dead, logs on nineteenth-century ancestors, photo records her aunt wanted shared. Locking those would have bought no safety and cost the ease of searching and reading her own notes. Per-entry encryption exists so the judgment can be made entry by entry — the sensitive minority protected properly, the working majority kept frictionless. She closed the weekend by copying the single vault file to her backup drive and opening it from the copy to prove it good. The tree program still held the tree. The vault now held the work.
Deadlines a Family Historian Actually Faces
Genealogy deadlines do not look like office deadlines, which is exactly why they slip. Nobody reminds Meg that the probate packet requested six weeks ago has not arrived, that the interlibrary-loan book is due back Thursday, or that the cousin who agreed to a DNA test has waited three weeks for the call about what the results might show. The cost of forgetting is quiet: a request that lapses, a loan recalled, a relative whose willingness cools. The calendar inside VaultBook, together with the Recent, Due, and Expiring sidebar, became the place she put those dates — not as a productivity system, but as a promise that the vault would show her, every time she opened it, what was recent, what was due, and what was about to expire.
Archive and courthouse requests earned the sidebar its keep first. Every time Meg sent a request — a probate packet from a county clerk, a deed copy from a recorder, a naturalisation record from a state archive — she created the request entry, attached a note of exactly what she had asked for and the fee sent, and set two dates: the date sent, and a follow-up date four or six weeks out, following the repository’s own turnaround guidance. The entry carried the status label waiting on record, so it sat in that Kanban column and surfaced in the sidebar as its follow-up date approached, with the next step already written in the log — the clerk’s name, the phone number, the reference number from the acknowledgement. Following up stopped being an act of memory and became an act of reading. When the packet arrived, the same entry received the result: date received, what was in it, what was missing.
Certificate orders worked the same way, with one addition: certificates fail instructively, and the vault keeps the instruction. A death certificate ordered with a ten-year search range that came back “not found” is evidence the death was not registered where she expected, or not in those years — and the failed order’s entry, with dates and details supplied, sits alongside the research log so she does not order the same search again in two years. Follow-up dates for certificates were shorter than for courthouse packets, because state offices either move or reject quickly, and a rejection sitting unopened in a pile of mail is weeks lost.
Interlibrary loan imposed the hardest deadlines, because the lending library set them. A borrowed county history that exists in only a handful of libraries arrives with a return date and sometimes a no-renewal condition, and the extraction — every Ellison and Hart mention in the index, the township maps, the church lists — must happen before it goes back. On arrival she created its entry, scanned the pages she needed under the library’s rules, and set the return date in the calendar with a working date a week earlier, leaving the final week as slack for the page she would inevitably realise she had missed. The Expiring view is built for this shape of deadline, and twice the early working date saved the work when the final week was consumed by family obligations.
The DNA calendar entries were different in kind, because they involved other people’s feelings. When a cousin agreed in principle to test, she created an entry recording the agreement and what had not yet been discussed — and set a date not for sending the kit, but for the consent conversation that had to come first. That conversation has a shape she follows and records: what the test can and cannot show, what surprises are genuinely possible — a parent not being the expected parent, a sibling relationship different from the one assumed — and what Meg will and will not do with the result, including that the match note will be encrypted, that nothing will be shared without a separate conversation, and that the cousin can withdraw and have their note deleted. Only after that conversation, recorded with its date and the cousin’s own words, did the kit get sent — with a follow-up date to check, gently, whether it had been returned to the lab. When results arrived, the entry closed and the match-note workflow took over, with the consent record attached permanently.
Family obligations got dates too, because they matter most and formalise least. Her aunt’s next photo session — an afternoon with the remaining shoeboxes, a scanner, and tea — went into the calendar the day it was agreed, with an entry listing which boxes, which photographs from the last session still needed names, and what to bring: enlarged prints of two photographs so her aunt’s eyes could work on faces comfortably, and the list of people in a group portrait a cousin had half-identified. Without the entry she would have arrived with the wrong box; with it, the session started where the last one ended. The same treatment covered a promised proof summary for a cousin, a drafted letter owed a final read, and the birthday of the oldest living relative in the tree — a date the vault could hold because the vault is where the family’s working memory lives.
Renewal dates for the services she used went into the calendar as dates, plainly, each with a note made at the last renewal recording what she had actually used it for: which record sets, which searches, which open questions still depended on it. When a date surfaced in the Expiring view, the decision in front of her was in her own words — this service holds the only online index to that county’s deeds and two open questions need it; or, this archive produced nothing in a year of monthly checks. The date is the fact the vault keeps. The judgment, made from the usage note rather than a last-minute scramble, is hers.
Repository trip planning is where calendar, labels, and offline design meet. A trip to a courthouse two counties away — or, on one occasion, to a basement records room with no mobile signal and a clerk who apologised for the flickering light — is planned in the vault weeks ahead. The trip entry gathers, by place label, every open question touching that county: the deed index years she needs, the probate packet too fragile to post, the marriage register page the online image cuts off at the margin. Before leaving, she confirms the vault file is current, checks the comparison images she will need are attached, and copies the file to a backup drive carried separately. In the basement, with no signal, the vault opens, searches, and shows her everything she brought. She photographs what the clerk brings, notes book and page numbers in the log on the spot, and when an unexpected deed names a brother she had never seen, she opens a new log entry at the table and files the photograph under it, because the vault asks nothing of a network. Back home, the trip entry gets its closing note: what was obtained, what the repository did not hold, what to request by post, and which follow-up dates to set.
The Weekly Routine That Keeps the Vault True
A vault does not stay true by being built well. It stays true by being tended in small, regular acts that prevent two failures: new material piling up unprocessed at the edges, and old conclusions sitting unrevised at the centre after the evidence under them has changed. Meg’s routine takes one long sitting — Sunday afternoon — and five habits inside it. Process one shoebox batch. Close one research log. Update one proof argument. Back up the file. Verify one record against its citation. Each attacks a specific way a vault goes stale.
Processing one shoebox batch comes first, while her attention is freshest. A batch is a deliberate unit: one envelope of photographs, one folder of letters, one bundle from a single source — not a whole box, which is a month of work disguised as an afternoon. She works it as she worked the original import, in miniature: each item scanned if needed, attached to its entry, identified as far as it can be that day, with identifications sourced and dated, and the physical item’s location recorded. Related notes do quiet work here: filing a photograph surfaces entries sharing its people and places, and twice a new photograph has contradicted an identification accepted from an earlier batch — a correction that only happens if new material is filed promptly. Unidentified items go into the photo ID workflow, and their questions go onto the Kanban board rather than into a mental pile labelled “later.” One batch a week is roughly fifty a year. The shoeboxes are finite.
Closing one research log comes second. Closing does not always mean answering the question; it means bringing one log to an honest state — work log current, sources cited as they will appear in the tree, findings separated from activity, and a decision recorded. Sometimes the question is answered, the entry moves by its status label to ready for tree, and its conclusion is entered in the tree program in the same sitting. Sometimes it cannot be answered with records currently reachable, and the log closes as a well-documented negative — searches made, repositories checked, records lost to a courthouse fire or registration gap — with next steps rewritten as the specific event that could reopen it: a record set not yet digitised, a trip not yet made, a DNA result not yet returned. Sometimes the question was simply the wrong question, framed around a man who turns out to be two men, and closing means writing that down and opening the right one. The Kanban board shows the effect: the open column does not grow without limit, because one card a week leaves it by design.
Updating one proof argument comes third, the most demanding habit, so it sits mid-sitting rather than at its tired end. Meg takes one argument in proof — often the one whose evidence has changed most — and rereads it against its template. Has a record arrived that strengthens or undermines a section? Does the conflicts section still name every disagreement she now knows, or has she been avoiding a birth-year discrepancy because the rest is tidy? Does the what-would-change-my-mind line still name the right falsifier — and has that record already arrived, filed under a different question? She searches the surname and place across everything, including OCR and indexed attachments, and uses the semantic question-and-answer over her own notes, because a vault of years contains connections its owner has forgotten she made. The argument is edited in place, its version history keeping earlier drafts, so if a conclusion later collapses she can find where a weak inference entered. An argument that reaches its standard joins the ready-for-tree work; one that falls apart is demoted honestly, its conclusion rewritten as a hypothesis.
Backing up the vault is fourth, and deliberately the least interesting habit, because a backup that requires thought will eventually be skipped. The backup is this: copy the single vault file to the backup drive, copy it to the second location she keeps, and open one copy to confirm it loads. That is the whole backup — no export, no selection, no separate attachments folder, no account. Entries, labels, attachments, encrypted entries, calendar and board all travel inside the one file. She keeps copies in dated folders, so she holds a shallow history of backups; more than once a copy from two months earlier answered what the current vault could not: what did this entry say before the new deed? Version history covers edits to individual entries; dated file copies cover everything else, including mistakes at the level of the whole vault. Backups fail in the gaps between systems. A single file has no gaps.
Verifying one downloaded record against its citation is last, smallest, and the conscience of the routine. Each week Meg picks one record — rotating through the direct line, favouring records her proof arguments lean on hardest — and checks it completely: open the image, read it in full rather than glancing at the line she originally needed, and compare it against its citation and transcription. Is the citation accurate in every element? Does the transcription preserve the original’s spelling, crossings-out, and marginal notes? Does the image match the citation, or is it a clear image of the wrong page, confidently filed? One record a week is fifty a year re-read by the person who relies on them. Errors surface steadily — a misread age, a witness’s surname one letter off, which in genealogy is a different family — but the main yield is familiarity. When a verification changes a fact, the entry is corrected the same day, the affected argument flagged, and, if the tree carries the wrong fact, the tree corrected with the vault entry as the record of why.
Around these habits, two tools shape the sessions themselves. The Kanban board holds research questions in four columns — open, waiting on records, proof in progress, ready to enter in the tree — mirroring the status labels, so board and labels never disagree. She reads it at the start of a session: what is waiting, and on whom; what is in proof and needs a clear head; what is ready and needs ten quiet minutes in the tree. And when a session ranges across several lines at once, dynamic tabs keep each line’s entries open side by side — a log, a photo record, a map entry — so she returns to each exactly where she left it. A multi-line session used to end with three half-finished threads; with tabs holding them apart, it ends with three shorter, finished pieces of work.
The routine is not sacred in its details. The day has moved twice, and the order bends when a loan is expiring or a cousin is visiting. What does not bend is the cadence and the smallness: one batch, one log, one argument, one backup, one verification.
What Changes for the Family
The first change the family noticed was a tree that got smaller, and better. For years Meg had kept one tree that tried to be everything — proven lines and promising leads, documented ancestors and strong suspicions, photographs everyone could see and stories not hers to tell. Sharing it meant sharing all of that, so she shared hesitantly, in fragments, with caveats nobody read. Once the working layer moved into the vault, the tree could be the clean, sourced, defensible account of the family, every claim in it able to survive “how do we know that?” because the proof work lived, finished, in the vault. From it she could produce a stripped-down public skeleton — names, dates, places, without working notes, hypotheses, DNA analysis, or interview passages given in confidence — and share it freely. A cousin in Oregon she had never met found the skeleton, recognised her own grandmother, and sent a photograph of a 1952 family gathering that filled a gap no Ohio relative could fill. That photograph arrived because the skeleton was safe to share widely. Separating the layers did not just protect the sensitive material; it liberated the shareable material, which had been held hostage by it.
The second change matters more: sensitive findings now move at human speed, under a protocol, instead of leaking through a sync. Every long-serving family historian finds something that rearranges a family’s understanding of itself — a parentage not as the birth certificate implies, a relative who disappeared from the records because they were in an institution the family never mentioned. Where notes sync automatically to shared trees, such a finding can propagate before its finder has finished checking it. Meg’s protocol runs in a fixed order. First, verify the finding to the proof standard, in the vault, encrypted, its what-would-change-my-mind line honestly answered — a painful claim that turns out wrong is a harm in itself. Second, identify the people most directly affected, living people first. Third, tell those people privately, in person or by phone, at the pace they set, with the understanding that it goes no further without their involvement. Fourth, only then consider the wider family and update the tree — or, in one case so far, deliberately not update the skeleton, while the documented truth stands, encrypted and complete, in the vault for a future generation to find proven rather than rumoured. A finding can be preserved without being published: neither broadcast nor destroyed. The family has understood that this discretion is not secrecy but sequencing, and it has made relatives more willing to tell her things — two recent interviews began, in effect: I will tell you this because I know where it will be kept, and where it will not.
The third change belongs to her aunt, whose knowledge is the archive no repository holds. Her identifications — names put to faces no one else alive could name, stories attached to objects, a family story told wrongly for forty years, corrected — are preserved in the vault with the photographs themselves, each dated, sourced to her, and recorded with her confidence and reasoning. That changes the sessions. They are no longer a race to extract information from a failing memory, conducted with dread; they are a collaboration with a record that accumulates. Each session starts from the last one’s entries, the enlargements prepared, the unnamed faces fewer, the aunt able to refine her own earlier identifications — “I told you that was Cousin Walter, but looking at the ear, that’s his brother” — with both versions kept and the change explained. When her health faltered for a season, the work paused without loss, because nothing essential lived only in Meg’s memory of conversations. The family understands what is being built: the testimony of the last person who knew, captured while she could give it, in a form still legible and checkable for a generation that never met her. Her aunt has begun bringing things to it, too — a letter kept separately for decades, produced with the instruction that it go in “the safe place,” meaning the vault, encrypted, with her name on the entry as its source.
The closing scene is a family reunion in a church hall in Ohio, of the kind the family has held for decades with better and worse potato salad. On the wall, Meg has pinned the printed chart from the tree program — the documented family across five generations, the Oregon branch added — and relatives stand before it finding themselves and their children, arguing happily about whether that birth year can be right, photographing the chart with their phones. Everything on that wall is there because it can be defended, and everyone is free to see all of it. On the table by Meg’s chair, her laptop sits closed. The vault is on it: the unproven argument she is still not satisfied with, the DNA notes under encryption and the consents that govern them, the photograph with two faces still unnamed, the interview passage her aunt gave her for the family but not for the wall, the finding that is preserved and proven and not yet, perhaps not in her lifetime, for sharing. It is in its place, the way private papers were always in a drawer nobody opened without asking. A cousin asks whether that is everything she has found. Meg says it is everything that is ready, and everything meant for the wall — and the rest is kept, properly, for when it is ready, or for whoever comes next and is ready for it. The chart is the family’s story as it can be told. The closed laptop is the family’s story as it is still being understood.
Start With the Shoebox
If you are where Meg was — a good tree, a decade of scattered files, a shoebox that matters, and a growing unease about where your sensitive notes actually live — do not start with the weekend. Start smaller than the weekend, because the smallest step is the one that will actually happen this week, and a vault begun is worth more than a vault perfectly planned.
Start with the shoebox. Choose one envelope — the one you would save from a fire — and give it the full treatment in a single sitting, in a new vault opened on your own machine. Scan its photographs if they are not scanned. Create a photo ID entry for each, attach the scan, and write down everything you already know: whose handwriting is on the back, who told you that was Uncle Ray, what you are only guessing. Label it with surname and place as best you know them, and mark its status honestly. Then create one research log entry for the single question that envelope raises most sharply, as one answerable sentence, with one concrete next step at the bottom. If anything in the envelope touches a living person’s privacy, encrypt that entry, choose a passphrase you will remember and keep safely, and reopen it to prove you can. Then copy the vault file to a second place.
What you will have is not a system. It is a demonstration, to yourself, of the difference between material that is merely saved and material that is kept — identified, sourced, findable, and held under your control. The labels for the rest of the family, the other templates, the audit of the decade of downloads, the courthouse trip planning: all of that can follow at the pace the weekly routine sets, one batch at a time, and it will follow more easily because the first envelope showed you the shape of the work.
That is the principle the whole of this article finally rests on, and it can be said plainly. Your family’s evidence should live in your custody, in a file you hold, that opens without anyone’s permission and answers to no server — because the work of understanding a family is too slow, too private, and too precious to be kept in a place you merely have access to. Custody is not a technical preference. For the person the family trusts with its shoeboxes and its secrets, custody is the job.