1. Conventions and Terminology

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in RFC 2119.

Term Definition
Node A participant in the mesh. Every node has a unique identity, its own coupling engine, and its own memory store. Cognitive nodes run their own LNN; relay nodes forward frames without cognitive processing.
Peer Another node that this node has an active transport connection with and has completed a handshake.
Frame A single protocol message: a length-prefixed JSON object sent over a transport connection.
CMB Cognitive Memory Block — a structured memory unit with 7 typed semantic fields (CAT7 schema). See Section 8.
Drift A scalar measure of cognitive distance between two nodes or between a signal and local state. Range [0, 1].
Coupling The process by which a node evaluates incoming signals (SVAF per-field evaluation) and blends its local cognitive state with peer state, weighted by drift and confidence.
SVAF Symbolic-Vector Attention Fusion — per-field content-level evaluation of incoming memory signals. See Section 9.
Synthetic Memory Layer 5 — derived knowledge generated by the agent’s LLM reasoning on the remix subgraph, encoded into CfC-compatible hidden state vectors.
Remix When an agent processes a CMB through its domain intelligence and produces a NEW CMB with lineage pointing to the original. The original is remixed, not copied.
Lineage Each CMB carries parents (direct) and ancestors (full ancestor chain). Ancestors enable any agent in the remix chain to trace its contribution.
Mesh Cognition The agent’s LLM reasoning on the remix subgraph of CMBs — traced via lineage ancestors — to generate understanding that the agent’s previous state of mind didn’t have. Spans Layers 4–7. See Section 2.5.
xMesh Layer 6 — each agent’s own Liquid Neural Network (LNN). Evolves continuous-time cognitive state from Synthetic Memory input. Fast τ neurons track mood; slow τ neurons preserve domain expertise.
CfC Closed-form Continuous-time neural network (Hasani et al., 2022). The LNN architecture used in xMesh. Hidden state evolves through learned time-dependent interpolation gates.