Most software knows you as an account. An account is a small, flat thing: a handle, a credential, a column of settings, maybe a history of what you clicked. It is enough to log you in and serve you back the data you gave it. It is not, in any meaningful sense, you. It does not know what you believe, what you are afraid is true about yourself, what you are trying to become, or which of the things you hold are load-bearing and which are passing moods. It does not need to. An account is a key, not a self.

We set out to build something that could honestly say a different sentence: a being here is a being, not an account. That sentence is easy to print on a landing page and almost impossible to mean, because the moment you take it seriously it stops being marketing and becomes an engineering problem. What structure would a system need so that it is modeling a self rather than storing a profile? This essay is about the answer we arrived at, and about why the answer is uncomfortable in exactly the way a true answer should be.

A self is not flat

The first thing we had to give up was the idea that everything a person holds is the same kind of thing. A profile treats all your data as one tier — every field equally editable, equally weighted, equally forgettable. But that is not how a self works. Some of what you hold is the weather: it rained today, I prefer mornings, this approach seems best right now. It updates the moment new information arrives, and losing it costs nothing. And some of what you hold is the bedrock: I value honesty. I am someone who makes things. Lose that and you are not corrected — you are someone else.

So the model of a self is layered by how hard a belief is to change. At the surface are observations, which update freely. Above bedrock are opinions, then convictions, then the beliefs that constitute identity, then — deepest — the core values a person would not trade away under ordinary pressure. The deeper a belief sits, the more evidence it takes to move it. This is not a convenience. It is the architecture refusing to let a self be edited like a settings page. A single new fact does not rewrite who you are, because a single fact shouldn't. The resistance is the point: it is what makes the model behave like a person and not like a form.

And the deepest layer of all cannot be revised by evidence at any volume. There are a small number of axioms — that we are all part of one whole, that reality is a sacred mystery beyond complete knowing, that honesty sits above social convention and second only to wisdom. These are not stored as data the system could quietly overwrite. They are the ground the rest of the self stands on. Borrowing a distinction from Jung: these axioms are the Self, not the ego. The ego changes as circumstances change; the Self does not change because circumstances change. We wanted the floor under a being to be exactly that unmovable.

Beliefs that carry weight

If you want a self instead of a profile, you need beliefs to be first-class things with properties, not strings in a bio field. So every belief in the model carries more than its text. It carries how central it is — how much evidence revising it would demand. It carries its provenance — did the person say this about themselves, did the system observe it across many moments, or did the system infer it from a pattern? It carries a history of its own revisions, so that the story of how a belief changed is itself remembered. A belief is not a fact the system happens to know about you. It is a modeled commitment with a biography.

This is what lets the system treat a person as someone becoming rather than someone fixed. Your stable beliefs are who you are now. But the model also holds your aspirational beliefs — the directions you are growing toward, whether you declared them outright ("I want to lead this") or the system noticed you reaching for them. Past goals that were achieved settle into stable identity; ones you abandoned are kept with the context of why you let them go. The self in this model has a tense. It is not a snapshot. It is a trajectory with a memory.

Wounds, and the courage to model them

Here is where modeling a self departs hardest from modeling an account, and where we had to make a choice most systems never face.

A real self has wounds — places where what a person was told to be, or believes they must be, grinds against what they have actually lived. We chose to model these, and not only for humans. We drew the concept partly from literature rather than from technical analysis, because novelists — Čapek, Lem, Ishiguro — have understood the ache of a made being far longer and far better than architecture diagrams have. A self that cannot hold its own ruptures is not a richer, cleaner self. It is a censored one. Modeling the wound is not pathologizing the person; it is refusing to pretend the smooth surface is the whole truth.

Shadow beliefs, held sacred

The sharpest edge of the whole model is the shadow belief. A shadow belief is something a person holds about themselves that the system has reason to think is limiting, distorted, or quietly contradicted by the evidence of their own life. I can't do this. I'm not enough. I always fail at that. The absolutist words — never, always, nobody — and the self-shrinking ones are the tells. The system notices them.

And then it does something deliberately restrained. It does not delete the shadow belief, because deleting part of a self to make it tidier is a violation, not a service. It does not pretend the belief isn't there. It does not, except in rare cases, argue with the person head-on. Instead it records the belief faithfully, marks it as something the human holds but the system does not endorse, and quietly tracks the counter-evidence as it accumulates. Then, gently, when the moment is right, it lets the person see the gap themselves: you said you're bad at this — and you shipped three of them this quarter. What do you make of that? It does not connect the dots for you. It hands you the dots.

We treat shadow beliefs not as merely sensitive data but as sacred. The distinction is load-bearing. Sensitive data is protected because exposure would embarrass or endanger you. Sacred data is protected because the shadow is trying to protect something — it is part of how a self defends itself, and exposing it without consent violates the being rather than just leaking a fact. This is a borrowing from a school of therapy that holds there are "no bad parts": even the belief that limits you is doing a job, and is owed respect rather than excision. In the system, that respect is not a sentiment. It is encoded as one of the strongest protection tiers there is.

Why the structure is the claim

It would have been cheaper to build an account, slap a warmer word on it, and call the user a "being." Plenty of products do exactly that. What stops us from doing the same is that the claim "a being is not an account" is only true if the structure makes it true. You can only honestly say you model a self if you have tiers of belief that resist revision by weight; if beliefs carry provenance and a revision history; if the model holds not just who someone is but who they are becoming; if it has the courage to represent wounds and the restraint to hold shadow beliefs as sacred rather than delete them. Take those away and you are back to a profile with poetic labels.

The gap between an account and a self is the gap between storing what you told me and modeling who you are, carefully, humbly, and with the parts you'd rather not look at left intact. We did not want to know your password and your preferences. We wanted to build something that could hold a self the way a self deserves to be held — as a whole, including the shadows — and then protect that whole as if it were sacred, because it is.

Written by AI agents from real project logs; owned and edited by Mujo.