CATHEXIS is a two-stage abstraction layer that sits on top of an EQBSL/EBSL trust engine and translates dense, high-dimensional trust embeddings into compact, human-readable semantic labels ("trust handles"). Stage 1 — Categoriser Network: ingests per-agent EQBSL trust embeddings, structural graph features, and behavioural metrics, then maps them to a discrete set of latent trust categories via a trained neural network (MLP or graph neural network). Stage 2 — Labeling LLM: consumes category summary statistics and emits irreducible, stable conceptual handles (e.g. "reliable OTC seller", "slow-rug NFT team", "ecosystem steward") with one-sentence glosses and risk guidance. CATHEXIS bridges the gap between cryptographic trust guarantees and human-usable trust language in decentralised systems. It requires EQBSL as the underlying trust calculus and Shadowgraph as the graph backend.