RouteGuard: Certifying Routing Gain in LLM Multi-Agent Systems When Complementarity Is Not Enough

This document is an arXiv statistical machine learning feed containing new research on LLM multi-agent routing certification, stochastic optimization, generative-model-assisted Monte Carlo sampling, conformal calibration and changepoint localization, interpretable spectral neurons, diffusion-based instrumental-variable regression, infinite-dimensional HMC, multi-kernel spectral clustering, and counterfactually fair reinforcement learning. The material is academic and does not describe software vulnerabilities, exploits, incidents, or active threats.