Internet of Things (IoT) agents that trigger network enforcement actions must be both well-calibrated (for safe triage) and tail-latency predictable (for service level objectives, SLOs). We present Confidence-Calibrated HP-FedGAT-Trust-IBN, a federated, graph-attention architecture that closes the loop from IoMT sensing to SDN enforcement via parameter-efficient (LoRA/PEFT) updates ( \(\le 1.1\) MB/round), trust-weighted secure aggregation, and intent verification (IBN) triage. Evaluation follows a two-plane protocol: a learning plane with \(N=100\) simulated clients under a matched comparator harness (Graph-FL and uncertainty-aware FL baselines), and a serving plane that replays exported checkpoints on real edge devices (Raspberry Pi 5, Jetson Orin Nano, Intel NUC 11) and validates SLOs using hardware ECDFs and empirical \(p99\) . The model achieves high discrimination (ROC-AUC/PR-AUC \(\approx 0.97\) – \(0.98\) ) with improved calibration (low ECE) under the matched harness, while the serving loop satisfies the \(<100\) ms requirement by device-measured \(p99\) (e.g., enforcement \(p99=27.5\) ms, vs. \(39.2\) ms for an efficient-UQ baseline) and explicit compliance \(\text{Pr}(\text{SLI}\le T)\) . The latency decomposition includes all calibration costs and Monte-Carlo expectations ( \(K=64\) , with measured MC share reported), and security modes are quantified end-to-end: CKKS + SMPC adds device-measured \(\Delta p99\) and crypto-attributable Joules (e.g., \(+26\) ms and \(+5.8\) J/round on Raspberry Pi 5). Energy/round is measured on identical hardware and mapped to CO2e for carbon-aware selection of operating points.
