My Aunt Asked Me for a Picture

The family is the cell of society. Soviet school slogan, drilled into every child, including this one

My aunt asked me for a picture of our family tree for her book. My daughter maintains our family tree on MyHeritage and the web site family tree rendering was not exactly what you’d want to add to a book. I already had a family model and a demo as well as a way to generate sites from diagrams - the building blocks were there. So my plan was to export our family tree from MyHeritage, and generate a diagram from it, optionally mapped to the family model.

I exported our family tree to a GEDCOM file and discovered that photos were useless - they were pointing to the cloud cache URLs which returned 403. Gemini told me that installing their desktop app would solve the image export problem - and it did. It also sovled the family tree picture problem because the desktop app produces decent family tree pictures, better than the web site. So, I was happy that I was done with it.

I was not done, because I have an infestation. Russian has a phrase for the private fixations that make a person tick: cockroaches in the head. The nearest English I have found is gremlins, so gremlins it is. Gremlinus cerebralis is a cogitavorous species: it feeds on unattended thoughts and nibbles on anything it can get hold of. It is also otiosurnal, most active during idle time - in the car and on the treadmill, on a trail.

A woodworker asked for one picture frame does not just cut four pieces of wood; she builds a jig, because the second frame is then free, and the tenth is identical. There is an old parable about the same instinct: two men are paid to fetch water, one hauls buckets all his life, the other spends his evenings building a pipeline. I have been building pipelines out of buckets my whole career, and the gremlins are why. My aunt had her picture. The gremlins wanted the jig.

GEDCOM has been the lingua franca of genealogy since the early nineties; every vendor imports and exports it, which means every family tree held hostage by a subscription is one export away from being data you actually own. I asked Claude to generate me an Xcore model based on GEDCOM and I already had an Xcore model archetype generating projects for Xcore models and those projects generate documentation sites and diagrams.

My first design was modest, and it was wrong in an instructive way. I have been building a tower of micro-models, each floor one concern: ownership, access control, lifecycle, decisions, governance, work, architecture, threat, and above the spine, capabilities, product management, organization design. The plan was to put genealogy on the lowest floor, the bare base, and let it climb as the tower solidified. Start small, add aspects later. Sensible, incremental, and it missed the point entirely.

The point arrived on the treadmill, delivered by gremlins. They kept listing what I actually wanted the model to do: track who in the family knows what, so that when my kids look for jobs in their fields we, as parents, can actually help with ours. Attach research tasks to the people they are about. Record decisions the way families actually make them. Keep the family’s ledger, because a family has one whether or not anyone writes it down, and the accounting floor of the tower fits a household without a single change, and not only for money: Covey called it the emotional bank account, politicians call it political capital, and The Godfather opens with a man booking a favor to be called in on a day that may never come. Accounts denominated in favors are accounts. Somewhere in that list the gremlins observed that I was not describing a genealogy database. I was describing an organization. They added that corporations routinely declare modeling too complex and let tribal knowledge rule, and that the phrase gives the game away. A tribe is a kinship organization, and tribal knowledge is what fits in the heads of people who all know each other, which caps it at the number of relationships a head can hold; Dunbar said about 150, and the critics argue about the number, not the cap. Families have modeled themselves for millennia precisely to get past it: genealogies are among the oldest records we have, and real tribes had a designated keeper of them, which is a higher standard than most corporations clear. A family model is complex, no doubt. Perhaps some families found it too complex, which would explain why they are not in the history books. I took the claim to an AI assistant, which is where I argue with gremlins these days, and the claim survived the argument.

A family is an organization. Not metaphorically: literally, in the modeling sense. It has members with capabilities, concerns it needs resolved, decisions it makes and revisits, work it needs done, events it remembers, resources it owns, and things it would rather keep private. Every one of those concerns already had a floor in the tower. So the model did not belong at the bottom, acquiring aspects over the years. It belonged at the very top, where all of them compose.

If “literally” sounds like a stretch, look at who some of the organizations are. Samsung is a family business, and so are Ford, Mars, Cargill and Hermès; the Rothschilds ran a banking network out of five cities through five sons before anyone had drawn an org chart. Even the fiction knows it: Dallas, which I know from an ABBA song and exactly one episode, ran for fourteen seasons on nothing but a family business behaving like one, succession, feuds, and wrath on every tie. What separates a family from a company is not structure. It is horizon. A company plans in quarters and, on a good day, in years. A family that intends to remain one plans in generations, and the proverb about shirtsleeves to shirtsleeves in three generations records what happens to the ones that do not. Most families never write anything down, which is precisely why the few that did are the ones we can name.

And at the top, something interesting, but not unexpected happened: the model nearly vanished. A family member is a person from the organization model, which already makes them a capability provider and a persona with concerns. Kinship ties are typed relationships from the architecture model; parent and spouse, biological and adoptive, are catalog entries, not classes. Events are named temporals from the core, which already knows how to say “about 1854” and “between 1912 and 1914”, because genealogy is where approximate dates live. What remains is the kinship vocabulary, GEDCOM round-tripping, and the thing genealogy got right two centuries before software: source citations with quality ratings. Which parish register says so, on which page, and how much the researcher trusted it.

Some things moved down, too. The tower plays Tetris: a concept the family model needs, once validated by centuries of use, drops to the floor where it belongs and clears a row for every model above. Architecture elements used to have a time coordinate; now they have a place and a jurisdiction as well, because a family model has to record that a great-grandmother lived in one house through three countries without moving. And the top floor comes with a test load: two centuries of GEDCOM, thousands of entities per file, every relationship type genealogy ever invented, free.

The renamed model is called family, not genealogy, because genealogy is one concern among many. Who is related to whom is the historical layer. Who can do what, and who can help whom, is the living one. A family tree that knows the uncle spent twenty years in security operations and the grandmother’s cousin sits on a hospital board is not a tree anymore; it is a capability network, and the questions it can answer (who can mentor the graduate, what careers have succeeded in this family, what is missing for the venture the cousins keep discussing) are questions no genealogy site and no professional network asks, because each of them holds only half the graph.

My aunt, it turned out, was not done. The picture was the first ask. The second was harder: would I write my part of the family book. Every family that keeps a book asks this of every generation, and every generation puts it off, because a life story written from memory is a blank page with a deadline. Written from the graph, it is a query: the events with their dates and places, the residences and occupations, the people who were there, the sources that say so, and, if I let the capability floor speak, what I turned out to be good at and whom I learned it from. That is a chapter. Narrow the same query to one purpose and one audience, and it is a CV. A resume is a chapter of the family book written for a stranger. Neither is the source of truth for the other; both are views.

The decisions matter as much as the capabilities. Families decide constantly: how to pay for college, whether the parent moves in or into assisted living, whether to buy the house. Some decide early. There is an old joke about a mother walking two small boys who is asked how old they are: the doctor is six, and the lawyer is four. It is a joke about a binding made at four and never revisited. The pattern I want is not a recommendation engine; it is analysis and binding. The computer lays out the alternatives with their consequences; the family chooses; the choice unlocks the next layer of decisions; and when reality intervenes (the scholarship falls through), the binding is revisited and everything downstream recomputes. The plan is not a document. It is a living graph that grows one human decision at a time, and the lawyer gets a say at fifteen.

Some of the graph must stay dark, and that is a feature. Hereditary health patterns and family finances belong in protected overlays that compose with the tree without living in it. The agent researching newspapers never sees genetic markers; the relative browsing the public site sees only the deceased. One model, several audiences, each seeing exactly what it should.

For the public demo I could not use my family; that would be a strange way to demonstrate privacy. The alternative arrived at the movies. I watched The Odyssey, and it sent me back to Nikolai Kun’s Legends and Myths of Ancient Greece, the book I read in my childhood. On the drive home the gremlins had a field day. Greek mythology is public knowledge, more or less well known, and nobody on Olympus has an expectation of privacy. A myth is a citation, and Hesiod and Homer disagree about Aphrodite’s parentage, which is exactly what citation quality ratings are for. Poseidon’s wrath at Odysseus is a dated valence on a tie, with the blinding of Polyphemus as the citing event. Gods and mortals want different shapes on the graph, and the graph wants a time slider, because the Titans and the heroes are not contemporaries. And the CV generator can be tested before it touches my own record. Odysseus, every capability backed by a source: strategy, the horse; carpentry, the bed; crew retention. Or Heracles, tailored to one specific posting: one day, one stable complex, three thousand cattle, thirty years of neglect, with a note on the pay dispute that followed. By the time I parked, the demo had a backlog.

The graph is the part I expect to matter beyond fun. Some time ago I built a 3D graph viewer, rotation, zoom, the works, to look at several thousand interdependent systems, and the clusters were visible before any analysis ran, the way a chemist sees a molecule’s shape before reading its formula. A family is a molecule. Force-directed layouts already compute the picture from forces on edges; give valence a sign and a strength and the picture does the field theory for you: the fused branch collapses into a knot, the estranged cousin drifts to the edge, and the relative everyone owes sits in the middle with every bond pulling on him. That is not decoration. Families have always navigated politics by computing those forces in their heads. Dune is the fictional extreme, ninety generations of marriages planned to steer power, and every real dynasty ran a smaller version. And where there is a ledger of favors there is concentration to measure, the way governance measures risk concentration: who holds the favors, and what changes when they are all spent at once.

Here is the part I did not expect to care about. I have spent years trying to interest organizations in composable models, and it is hard, because in an organization it is not people’s own money and not people’s own inefficiencies. In a family it is both. There are millions of developers and tech-savvy people with families, and every one of them owns exactly one dataset they are guaranteed to still care about in thirty years. The family model sits at the top of the tower, which means it quietly exercises every floor below it. It is the showcase that fits in your pocket, and how organizations eventually get interested in towers.

Thirty years is not a figure of speech. I still have files I wrote thirty years ago, and I can open them, which is more than I can say for most of the products that wrote them. Name a vendor you would trust with your family’s record for the next thirty. Not with the data; clouds rarely lose data. With the product. Products get acquired, sunset, repriced and redesigned. So the trust posture here is a survivability test, and I apply it to Nasdanika itself. Suppose Nasdanika disappears. The model files are plain text: Xcore, Draw.io XML, Groovy DSL, and GEDCOM itself, readable in any editor and diffable in Git. The released artifacts stay on Maven Central, which does not delete what has been published. The command-line distribution bundles every dependency down to the JVM, a runtime that is itself past thirty and still executes class files compiled in the nineties. Everything a family can do with its model the day Nasdanika stops, it can do the day after, from a flash drive in a safe. I trust that flash drive over any cloud vendor, Microsoft, Amazon and Google included. Not because I have evidence they will lose my files, but because I have no evidence they will still be offering this particular product, on these terms, when my grandchildren need it.

That is also where the model sits in the landscape, and it is worth saying plainly. The industry is built for huge volumes of simple records that matter for a while: clicks, transactions, telemetry. A family’s record is the opposite corner: thousands of entities, not billions, every one of them tangled in relationships with the others, and mattering for a century. Small, complex and permanent is a corner the large platforms do not optimize for, and it is the corner Nasdanika was built to occupy.

My aunt got her picture. It is a better picture than the one I did not screenshot, but that is not the point. The next ask, the next correction, my chapter of the book, and the branch my cousin wants to maintain herself are all regenerations now. Haul a bucket when someone is thirsty. If you notice the asks repeating, build the pipeline. If the pipeline turns out to be an organization, put it where organizations go: at the top. And if you have gremlins, put them on a treadmill.