Executor-Side Progressive Risk-Gated Actuation for Agentic AI in Wireless Supervisory Control
The PRGA method structures AI control intents into three evidence layers: C0 for local triage, C1 for on-demand coordination evidence, and C2 for post-hoc provenance. This approach improves performance by reducing time to first safe action by 23.3–27.4% and cutting control-plane bytes by over 50% compared to full-evidence methods while maintaining safety margins.