Calibrated-Uncertainty Detection for Counter-UAS: Order-Invariant Evidential Fusion with Guaranteed-Containment Tracking

AEGIS: Calibrated-Uncertainty Detection for Counter-UAS AEGIS is a detection-only, human-in-the-loop counter-UAS (C-UAS) software pipeline built around calibrated, honest uncertainty — a system designed to know when it does not know. It discriminates drones from birds, clutter and empty scenes under noisy and sometimes contradictory sensor evidence, and is engineered to surface disagreement as calibrated uncertainty and to abstain under genuine conflict rather than emit a confident wrong answer. AEGIS performs detection, classification, tracking and alerting only; a human makes every decision, and the system has no countermeasure or targeting capability. The pipeline combines four components: order-invariant n-ary evidential fusion (PCR6) that removes the order dependence of iterated pairwise combination and retains more declared ignorance under multi-source conflict at equal discrimination; guaranteed-containment set-membership (zonotope) tracking whose position-containment guarantee is validated against real UAV flight trajectories, including aggressive fixed-wing manoeuvres that violate the tracker's own acceleration assumptions in 10–13% of ticks; runtime verification of bounded temporal-logic safety properties; and hardened calibration metrics with an explicit abstention outcome. Every run is deterministic under a fixed seed and is written to a canonicalised, hashable audit trail, making results bit-reproducible and independently checkable across platforms. This deposit contains the reference implementation, the reproducible synthetic testbed, the real-flight validation experiments, and the documentation, including an honest account of the system's measured limits (data-independent set width, position-only consistency monitor, weakly observable velocity) and the sensor that would close them. Repository: https://github.com/QuantuSync/AEGISLicense: Apache-2.0Author: Lucas Alaniz Pintos (Gerencia de Smart Products, INECO, Madrid, Spain), ORCID 0009-0008-5179-2534