Roadblock: Bounding Data Exposure in Grid-Edge IoT Networks with Subnet Isolation and Decision-Only Reporting

Grid-edge IoT deployments increasingly combine heterogeneous sensing, local inference, and externally reachable reporting interfaces. These systems improve situational awareness but create privacy and security risks when raw or intermediate sensor representations persist or leave the components that need them. This paper presents , an edge architecture that uses Snap IoT as its topology-control engine. Snap IoT pools devices by sensor type behind authoritative devices (ADs) and constrains authorized communication paths; Roadblock adds gateway curation, discard-at-edge processing, and decision-only reporting. In every Roadblock-conforming Snap IoT topology, devices follow the same role sequence—device to AD to middle tier (MT) to decision reporting layer (DRP)—regardless of the number or arrangement of devices and pools. Generated raw and intermediate IoT data terminate within internal processing, while only compact 32-bit decisions are eligible for external export. We implement one conforming topology in a laboratory grid-edge facility testbed with camera, audio, temperature, and device-state subnets. Correlated encrypted packet captures and validation logs show no observed unintended cross-subnet traffic, no raw sensor outputs submitted to the reporting path, and preserved fault detection, environmental detection, and remediation workflows. Attack-driven runs further test cross-subnet reachability and reporting-plane inbound access. The deployment validates the implementation; containment across conforming topologies follows from the role-constrained construction.