SCALER: Self-Consistent Assessment for SLA Explanation and Resolution in 6G Networks

In emerging 5G and beyond (B5G) network archi tectures, ensuring service-level objectives (SLOs) requires more than reactive monitoring. It demands systems capable of inter preting behavioral deviations and reasoning about compliance with operational intent. This paper introduces a framework that combines unsupervised behavior modeling with large language model (LLM)–driven reasoning to explain potential service level agreement (SLA) non-compliance. By leveraging a multi-sample zero-shot prompting strategy to generate contextual explanations for observed service degradations, the system aligns intent ex pression with network observations. To assess the reliability of these explanations, it introduces a self-consistency mechanism based on semantic similarity among multiple LLM responses. This mechanism acts as both a proxy for explanation confidence and a signal for semantic drift. The architecture is deployed in a cloud native pipeline, in line with zero-touch service management and scalable orchestration. Evaluations on real-world traffic datasets demonstrate the system’s ability to support trustworthy and interpretable SLA assurance in machine learning (ML)-based telecommunication environments.