Nothing matches that filter.
The first plateau is the baseline picture Priostack holds of a person — who they are, what they carry, what they will not have automated — authored as a real model before the assistant does anything on their behalf. This spec defines what it contains, how it is captured with consent, when it is complete enough, and how Γ reads it.
It is for anyone, anywhere. The hundred things below are human universals, phrased to hold across cultures, faiths, incomes, family shapes, and abilities — never a data-broker profile.
Priostack acts on a person’s behalf: it prioritizes, decides, and sometimes moves money or reaches people. It can only do that with dignity if it holds a true model of the person — their environment, obligations, joys, pain, struggle, and above all the lines they do not want crossed.
The first plateau is that model at its starting, coherent state. It exists so decisions are made against a real world, not a blank one, and so Γ — the engine’s appraisal language — has a truthful baseline to read against. Γ can only tell drift from normal, or novelty from routine, if “normal” is richly and honestly described first.
Priostack runs one chain from the model of the person down to the behaviour they experience. The plateau is where it starts: the ArchiMate model. Everything downstream reads from it.
Each knowing is placed on the plateau layer it belongs to. That placement is not decoration: it decides how the knowing is used downstream — a motivation knowing shapes what the assistant optimizes for, a constraint becomes a hard gate in a decision table, a context knowing colours how Γ reads a situation.
Distribution of the hundred knowings across plateau layers.
A knowing is never just a value. It is stored with a signal type that tells the substrate how to treat it, a source and confidence that say how much to trust it, and a freshness stamp so stale facts decay rather than mislead. The signal type is the load-bearing field:
| Signal type | What it is | How the system treats it |
|---|---|---|
| stable-fact | Rarely changes (identity, origin, a chronic condition). | Trusted long; a change is itself notable and re-confirmed, not silently overwritten. |
| changing-state | Expected to move (income, health, mood, location). | Its trajectory is exactly what Γ watches — drift here is signal, not error. |
| preference | How the person wants to be treated (tone, channel, timing). | Shapes behaviour and voice; never treated as a fact about the world. |
| constraint | A hard line (do-not-disturb, never automate this, a limit). | A gate on decisions. Enforced, never traded away for convenience or by Γ. |
| relationship | A tie to a person, group, org, or place. | An edge in the model; carries its own consent and its own priority. |
Each knowing records how it was learned — ask (the person told us), observe (from their own activity), app (a service they use, shared with consent), or derive (inferred, only where inference was consented). Derived knowings carry lower confidence and are the first to be re-confirmed.
No one is asked a hundred questions. The first plateau is seeded by a short, tone-adaptive conversation that captures only the load-bearing knowings, then grows over time from what the person chooses to share, what their apps contribute with consent, and what can be safely derived.
Consent is per-knowing and purpose-bound. Sensitive knowings (health, faith, finances, relationships, safety) require an explicit basis and are stored at a higher protection tier. The person can decline, redact, or correct anything; a redaction is honoured downstream, not merely hidden.
The strongest signal of respect is recording what the person does not want automated, known, or shared — and when never to disturb them. Those are first-class constraints in the plateau, and they outrank convenience every time.
A first plateau is solid when each layer holds enough true knowings for Γ to have a baseline, and the load-bearing minimum is present. Until then, Γ honestly reports UNKNOWN rather than guessing — the same way the engine returns “no evidence” against a thin baseline rather than inventing a verdict. Completeness is a coverage bar per layer, not a checklist score.
The minimum for Γ to be honest — the knowings without which the assistant should not act unsupervised:
Every decision the person makes runs on the substrate and leaves a trajectory in geometric memory. Γ compares each new trajectory against the baseline the plateau establishes. It narrates only on a genuine deviation — something outside known territory, or a state that has really begun to migrate — and stays quiet the rest of the time. Constraints are not appraised; they are enforced.
This is why the plateau must be rich and true: a thin or wrong baseline makes Γ either cry wolf or miss the moment that mattered. A good plateau means fewer false alarms and more true ones.
“You send help to family most months, so this month’s transfer is expected — I won’t flag it. But it went to someone new and larger than you ever send, so I’ll check it’s really you.”
Lifecycle. The first plateau is the baseline. Life events — a new child, a move, a loss, a job change — author successive plateaus, each versioned. The transition between two plateaus is a real object Γ can narrate. Stale knowings decay in confidence and are re-confirmed rather than assumed.
Representation. The plateau is authored as a real ArchiMate model on the substrate — not a form’s worth of fields. Each knowing is an element on its layer; relationships are edges; the whole is versioned and is what decisions (BPMN/DMN), the interface (IFML), and Γ all read from.
Not surveillance, not a dossier, not a personality score. Measure, don’t assert. UNKNOWN over fabrication. The plateau serves the person and is owned by them; every knowing is true, uncertain-and-marked, or absent — never guessed.
The concrete content: a hundred universal things worth knowing, each mapped to its plateau layer, signal type, and source, with an example of the Γ narration it enables. Filter by layer or search to explore.