Reception Infrastructure for Conversational Agents: Consent, Memory Provenance, and Runtime Governance in Local LLM Systems

We present a reproducible runtime governance architecture for local conversationalagents. The system combines consent scaffolding, memory provenance layers, personaanchoring, pressure and drift telemetry, and reflective session logging around a locallydeployed open-weight language model. By reception infrastructure, we mean runtimesystems that prepare, constrain, observe, and preserve the conditions under which aconversational agent receives memory, identity-relevant context, user interaction, andcorrective feedback. This differs from output containment or post-hoc moderation. Weevaluate the architecture across governed and ungoverned baseline conditions, afour-condition controlled doctrine encoding experiment, adversarial sessions, andguided cultural-literacy sessions. We don't claim consciousness, sentience, orpersonhood. We examine whether separating authority, memory, persona, telemetry,and reflective continuity into auditable layers improves conversational stability,provenance-aware self-reference, and recovery from pressure events. Implementationis available under MIT license at github.com/antibox-riot/synthetic-life-charter.