Social engine
The subsystem that maintains persistent models of users and peers, infers intent and belief states, and tracks trust and deception risk through separate update mechanisms.
The social engine extends the memory substrate to model not just operational history but the agents - human and otherwise - that the system interacts with. For each persistent user or peer, it maintains a belief model: what does this agent appear to believe, what are their likely goals, how consistently have their stated intentions matched their observed behavior? Trust and deception risk are tracked through separate update mechanisms because they are not simply inverses of each other. Trust rises through consistent, accurate, beneficial interaction; it falls when predictions about a peer's behavior are wrong or when interactions produce poor outcomes. Deception risk rises when behavioral signals are inconsistent with stated intent or when an interaction pattern matches known manipulation signatures; it is specifically watched separately so that a sequence of mostly-trustworthy interactions does not mask a developing deception pattern. The social engine feeds its user and peer models into the agent context at session hydration, giving the planner and critic access to the system's current assessment of the counterparty when forming strategy and risk estimates.