ARCANA: Autonomy Risk Calculus for Agentic Network Assurance

ARCANA (Autonomy Risk Calculus for Agentic Network Assurance) is an open research and reference framework for model-bounded autonomy accounting in agentic systems. It represents capability propagation through a horizon-bound nonnegative graph model, separates unsafe propagation pressure from loss or impact, carries calibration uncertainty explicitly, and emits typed risk artifacts with granular reason codes. The accompanying reference implementation and ARCANA-Bench evidence use deterministic synthetic fixtures. This preprint does not claim absolute safety, external peer review, empirical deployment validation, production enforcement, or production certificate authority. ARCANA remains independently usable; production-grade integration with AXCP Enterprise/CBE is outside this public record. Repository: https://github.com/tradephantom/arcana. Related AXCP project: https://getaxcp.com.