Rater State Bias in RLHF Preference Data: An Audit Framework When Does Muon Help Agentic Reinforcement Learning? Kernelized Linear Attention: Breaking the Capacity Wall with Symmetric Cones CoEvoP&R: Co-Evolving Placement Objectives with Routing Feedback via Large Language Models Induction in Both Directions: A Mechanistic Analysis of In-Context Learning in Masked Diffusion Language Models