AI-Driven Self-Healing Systems via Failure-Conditioned Reasoning Loops
Sahil Agarwal · Zenodo (CERN European Organization for Nuclear Research) · 2026
This paper introduces Failure-Conditioned Reasoning Loops (FCRLs), a closed-loop architecture for autonomous remediation in distributed systems. Unlike traditional trigger-action automation, FCRLs treat failed interventions and verifier diagnostics as first-class evidence that conditions subsequent decision-making. The paper formalizes loop dynamics, policy gating, verifier-gated termination, and safety constraints, and presents an evaluation framework for empirical validation. This work reframes self-healing from reactive execution to verifier-gated stabilization under bounded experimentation.