Cross-Model Friction (CMF): A Human-Arbitrated Protocol for Multi-Model Adversarial Refinement — Extended Abstract

Extended abstract presenting Cross-Model Friction (CMF), a methodology that treats independent Large Language Models as separate analytical laboratories. Unlike ensemble methods or multi-agent debate systems, CMF preserves complete model independence, requires a mandatory Human Anchor (Layer 0) for all synthesis decisions, and maintains an immutable Witness Layer via Knowledge Continuity Engineering (KCE). Validated on emerging profession forecasting (2026–2035) using 11 LLMs from 4 countries, producing a 77-profession taxonomy with per-entry model provenance. Full methodology and dataset available for licensing.