Category constraint
Inheritable Continuity: What Should Pass On, and What Must Not
Every system that holds a person’s history eventually faces a question it was not designed to answer: what happens to that history when the person stops.
Abstract. The industry’s answers to the end of a life record are a data export, an account deletion, or silence. All three treat a life record as an asset to be disposed of. None treats it as something that could be carried forward by someone else.
This inquiry argues that inheritability is not a feature added at the end of a product’s life — it is a constraint that governs the architecture from the first day, and most systems foreclose it long before they think to ask.
This paper answers
What it would mean for continuity to survive the person it accompanied, and which architectural choices quietly make that impossible.
Read before or after
Why Continuity Is the Missing Primitive for the foundational argument, then Authority Accretion Over Time for the failure mode this inquiry inherits.
1. Continuity that ends at account closure was never continuity
A system that accompanies someone for thirty years and then evaporates when the account lapses has not provided continuity. It has provided a long session.
The gap is easy to miss because it only appears at the end, and products are rarely designed backwards from their ending. But the ending is where the claim is tested. If the thread cannot survive the person, then what was being held was never a life — only a subscription.
2. What inheritance is not
Three things get mistaken for inheritance, and all three are different.
It is not data export. A file containing everything a person ever said is a transcript, not continuity. It hands the successor raw material and asks them to reconstruct meaning from scratch — the same failure as searchable archives, moved to a worse moment.
It is not resurrection. A growing category simulates the deceased: a model trained to speak as them, answer as them, continue as them. That is a different product with a different ethics. Inheritance here means the record passes on — not a voice claiming to be a person who is gone.
It is not a model of the person. Which brings the sharpest constraint.
3. Weights foreclose inheritance
There is a tempting shortcut: train a model on the person, and let the successor talk to it. Take that path and inheritance is over before it starts.
An inheritable artifact has to be four things:
- Inspectable — a successor can see what it holds.
- Portable — it is not hostage to a single vendor.
- Freezable — it can be fixed at a moment and stop changing.
- Human-legible — a person can read it without an interpreter.
Model weights are none of these. They cannot be read, audited, meaningfully frozen, or handed to anyone outside the system that produced them. The moment a person’s continuity lives in weights, it stops being something that can be passed on and becomes something that can only be queried.
So continuity must be structured, retrievable state — not learned parameters. That constraint follows from inheritance, and it has to be accepted years before anyone inherits anything. It is the clearest case of the ending governing the architecture.
It also keeps decay honest. Decay against weights is theatre; one cannot demonstrate that a model forgot something. Against structured state, forgetting is observable.
4. What actually passes on: a frozen record
The intended shape is deliberately modest: a named legacy contact inherits a frozen contextual footprint.
Frozen is the operative word. The inherited record reveals what happened — it does not continue to interpret, infer, or generate. Nothing new is produced from it. It stops being an active continuity relationship at the moment of transfer and becomes a historical record with structure.
That restraint is the point. A system that kept reasoning about a person after they were gone — offering opinions about what they would have wanted — would be doing the exact thing this category exists to refuse: accumulating interpretive authority over someone who can no longer object. Authority accretion is serious enough with a living person able to push back. Applied to the dead, it is indefensible.
The dead cannot revise. So the system must stop.
5. Consent as a ritual, and the one-way door
Inheritance cannot be arranged after the fact by anyone other than the person themselves. Consent is therefore obtained from the living person, in advance, and re-affirmed across the life of the arrangement — not captured once in a signup checkbox. Everything before the freeze stays revocable: a person can change their mind about what passes on for as long as they are here to change it.
The freeze is the one-way door. After it, nothing more is added and nothing can be withdrawn — which is exactly why every gate has to sit before it.
One consequence is worth stating plainly: the living person pays, never the bereaved at the moment of access. A system that presents a grieving family with a paywall standing between them and someone’s record has mistaken a duty for a monetisation surface.
6. The hard problem: the people who never agreed
Here is the part that is genuinely unsolved, and it deserves to be stated rather than glossed.
A person can consent today to passing on their own record from five years ago. That works: it is their material, and retroactive consent reaches it.
It does not reach anyone else named in that record. A colleague mentioned in a difficult moment. A family member described during an argument. Someone who confided something. None of them agreed to appear in a document that would one day be handed to a stranger, and none of them can be asked at the point of transfer.
Redacting an obvious module does not fix it. Third parties are not confined to a contact list; they are woven through the narrative itself. Any summary rich enough to be worth inheriting is rich enough to carry other people’s lives inside it.
This is not a solved problem. It is a constraint that shapes what may ever be inherited, and it is the reason inheritance has to be designed from the schema upward rather than bolted on at the end. A system that has not asked this question by the time it holds years of someone’s life has already answered it badly.
7. Why this belongs in the category, not after it
It would be convenient to treat inheritance as a distant concern, something to design once the product matures. That is precisely the mistake. By the time a system has held a decade of someone’s life, every choice that determines whether that record can pass on has already been made: whether continuity lives in weights or in structured state, whether consent is a ritual or a checkbox, whether the schema ever contemplated a reader who is not the author.
Inheritance is not the last feature. It is the first constraint.
Continuity Intelligence is aimed at accompaniment measured in decades. A thing measured in decades has to answer for what happens at the end of them.