Post Content Post navigation The Illusion of Procedural Reasoning: Measuring Long-Horizon FSM Execution in LLMs VLA-Pruner: Temporal-Aware Dual-Level Visual Token Pruning for Efficient Vision-Language-Action Inference