We address robust monitoring of distributed data streams where up to a fraction \(\beta < 1/2\) of n nodes may be Byzantine. For resilient, low-overhead aggregation, we propose a deterministic distributed algorithm computing a coordinate-wise \(\beta\) -trimmed mean via Q-Digest sketches. Nodes efficiently summarize local data, merged along a spanning tree into a robust global aggregator. An event-driven variant further cuts communication in stable periods using per-node safe intervals, triggering updates only on significant local deviations. We evaluate our algorithms using comprehensive simulations with realistic, adversarial data, comparing decentralized (periodic and event-driven) methods against a list aggregator. Our protocols achieve bounded aggregation error while substantially reducing communication, energy, and storage. Validation on MAWI network traffic (/24 subnets as virtual sensors with flow features) demonstrates both Q-Digest aggregators match centralized trimmed-mean accuracy amid bursty adversarial injections, while the event-driven variant cuts communication by over 75% in this high-rate scenario.
