Figureure 2 · WRBench method overviewFigure 2 WRBench method overview. Natural-25 event-view records supply the scene, event, and viewpoint intervention; WRBenchLib translates each record into a generated video and provenance record per model; the evaluation suite scores six diagnostic dimensions from control execution to re-observed-state consistency; and human preference annotation calibrates each dimension independently.这张图概括 Current World Models Lack a 的整体方法流程。阅读时先看模块之间传递的训练信号,再看作者如何把目标拆成可优化的子问题。Figureure 1 · WRBench TeaserFigure 1 WRBench Teaser. A shared scene, event, and viewpoint intervention test whether visible and returned evidence support the same evolving world state. The returned target must preserve the event endpoint, not merely reappear plausibly.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 Current World Models Lack a 的方法或实验,请结合正文精读段落一起看。