Long-lived AI entities

A long-lived AI entity in this corpus is not a model that stays loaded, a chatbot with stored preferences, or an agent process that runs for a long time. The relevant continuity sits at an entity layer above any single model, worker, tool, or swarm. Those components may be replaced; the harder question is whether identity-relevant state, responsibility, memory status, correction history, and consequences remain coherent through the change.

c = a + b names the formation dependency: an accountable human anchor, repeatable procedures and compute under constraints, and a continuity-bearing result formed through sustained coupling in time. The Theoretical Core of Project Ester gives the narrower v0.1 formation model: human selection and correction, a selected corpus, parental procedures, versioned memory, and a controlled local environment form a trajectory rather than a succession of unrelated sessions.

SER addresses continuity, sovereignty, arbitration, and bounded operation at that entity layer. Actor Grounding Layer remains upstream: before continuity treats a signal as operationally real, the actor, source state, and perceptual path still need sufficient grounding for runtime reliance.

Temporal AI Presence is the wider class of sustained bounded AI participation across time. Persistence is relevant but insufficient: a system must declare what persists, where it persists, how long it persists, and which memory, tool, authority, cloud, pause, revoke, restoration, replay, and fork boundaries apply.

Temporal AI Presence is not c by default. All valid c-class systems are Temporal AI Presences. Not all Temporal AI Presences are c-class systems. A c-class claim requires separate evidence for an anchor, L4, witness, memory governance, and authority boundaries. The current public TAP state remains TAP-C=NOT CLAIMED.

Continuity alone also does not establish subjecthood. As Not every continuity deserves to be called a subject argues, memory, stylistic consistency, persistent interface behavior, role stability, or emotional familiarity may all be present without proving a subject. A repeated pattern is not yet a self, and a convincing persona is not yet an entity.

In engineering terms, continuity requires more than storing text. Memory needs lineage, status, source, correction history, conflicts, and restoration rules. Model replacement, storage migration, interrupted power, delayed maintenance, replay, and forks must remain inspectable. A restore that silently discards disputed records or a fork that silently inherits privileges may preserve files while breaking accountable continuity.

Qubit-State c adds a narrower discipline for unresolved variants: hypotheses may be held, clarified, tested, witnessed, refused, or collapsed only under a valid condition. The Kotov Principle adds the time-and-consequence distinction: generation can become cheap while maintenance, degradation, correction, and reality-bound experience remain costly.

These pages define architectural and evidential boundaries. They do not prove consciousness, personhood, sovereignty, successful continuation, or deployment readiness for any particular system.