Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding Written on April 7, 2026