Field Guard: Empirical Validation of a Contraction-Based Stability Layer

Read full text
Authors: James Foster
Cited by: 0

Abstract

Finnish patent application no. 20265235, filed 11.03.2026, REAL E3 Systems Oy.Field Guard is a deterministic, mathematically certified stability layer for real-time AI and multi-agent systems. The formal mathematical certificate is established in WP-FG-002 (DOI: 10.5281/zenodo.18831894). This paper presents the empirical validation of that certificate across 13 independent harnesses. Validation covers four categories: mathematical verification of the certificate against the live codebase (H01–H04); adversarial and long-horizon LLM testing using self-assessment extraction (H05–H08); real-world human behaviour testing using an external wrapper with zero LLM modification (H09–H12); and a 24-agent national crisis simulation testing ISS certificate composition at multi-sector scale (H13). Key results: drift reduction ranged from 43.0% to 61.2% across all empirical harnesses. In no harness did the governed system exceed the certified ISS bound. Zero certificate violations. In the 24-agent crisis simulation, the governed system recovered to steady state within 1 turn after peak blackout stress; the ungoverned system did not recover within the simulation window. The Field Guard v2.0 implementation is 187 lines of executable Python, requires no GPU, and in external wrapper mode governs any target system with zero modification.

Fields of Study & Subjects

Formal Methods in Verification Adversarial Robustness in Machine Learning Safety Systems Engineering in Autonomy

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
4 March 2026
Language
EN