A MoE expert separation workbench for building, testing, and routing smaller specialist models with MetaHarness, Ruflo, and LatentMesh state exchange. Includes reproducible benchmarks, quality gates, architecture decisions, and animated explainers. - View it on GitHub