LLM-as-a-Judge: Model Routing and Scoring in GreyMatter

This report describes ReliaQuest GreyMatter's "LLM-as-a-Judge" evaluation layer: a model-agnostic system that routes tasks to appropriate model tiers, evaluates output quality against rubrics, and improves via user feedback and continuous monitoring. It explains decision logic for when to use frontier LLMs versus lightweight deterministic checks, safeguards to prevent the judge becoming a bottleneck, and operational recommendations for assessing platform quality and model usage.