Automatically judged by Luoshen and Qwen multimodal models, with public pre-game rules and explainable results.