Yuxin Meng, Yuhan Suo, Junjie Wang +9
Existing benchmarks for MLLM-generated web artifacts assess interaction through local evidence and miss the requirement-induced states and transitions that determine whether a page…