Evolvable Multi-Agent Hierarchical Architecture for Contested Logistics Ecosystems
6.3 ILLUSTRATIVE DECISION CYCLE
Under a disruption scenario combining infrastructure degradation with data anomalies indicative of potential manipulation, the architecture operates through a five-phase decision cycle:
• Sense, validate, and anomaly-detect( tactical level): Tactical agents ingest streaming updates submitted to the data platform. The platform checks completeness, timeliness, and consistency, attaching provenance and quality indicators. It also runs cross-source contradiction analysis: where independent sources should agree but diverge, anomaly flags are raised. Data patterns deviating from established norms, including patterns consistent with known adversarial attack signatures, [ 13 ] trigger adversarial validation signals for escalation rather than silent acceptance.
• Fuse and infer( platform and operational level): Operational agents fuse available data into a situational picture reflecting current state, uncertainty, and flagged anomalies. Capacity constraints are represented as confidence-bounded estimates, enabling explicit reasoning about risk, including risk arising from potentially adversarial inputs.
• Propose and coordinate( operational and strategic levels): Operational agents generate feasible candidate actions. Strategic agents evaluate proposals against higher-level objectives, risk tolerance, and flagged data concerns. The hierarchy supports iterative refinement rather than monolithic optimisation.
• Human-aligned decision and execution: Where decisions have high consequences or involve data anomalies, human decision-makers remain in the loop [ 12 ]. Agents provide recommendations, rationale, confidence indicators, and anomaly flags. Human oversight is particularly important when adversarial manipulation is suspected.
• Monitor, learn, and adapt: Performance signals and anomaly feedback are captured back into the shared platform, supporting refinement of agent policies, data quality rules, and adversarial detection thresholds. This directly realises the evolvable design intent.
6.4 BUSINESS OUTCOMES
Table 6-1 maps the primary business outcomes to enabling architectural features and indicative KPI dimensions, applicable across defence, government, and commercial supply chain settings.
EDM Association – Journal of Innovation 69