Operational Routing Intelligence (ORI): A Reference Architecture for Evidence-Based Adaptive Routing in Wholesale VoIP Networks

Abstract

Adaptive wholesale VoIP routing is often implemented as a direct mapping from quality metrics to route changes, leaving the operational justification implicit. This paper defines Operational Routing Intelligence (ORI), a reference architecture that separates observation, reliability, evidence accumulation, operational state, authorised action, execution and outcome validation. A Python prototype was evaluated in an audited synthetic benchmark with three supplier routes, eight scenarios, four routing policies, two information modes and 500 untouched hold-out seeds. Under equal information, ORI detected sustained normal-volume degradation in every run, with a conditional median delay of 5 min and approximately one material policy change per event. Its behaviour was more conservative under weak or conflicting evidence: detection probability fell to 0.49 under sparse traffic, and the equal-information policy did not detect the mixed asynchronous failure. Full information raised detection probability in that scenario to 0.152. ORI also averaged 1.18 false interventions in the transient scenario and did not complete validated recovery. The benchmark therefore shows that explicit operational reasoning is implementable and traceable, but does not establish universal superiority or production readiness.