Latent Structure Benchmark — applies Cultural Domain Analysis elicitation protocols to large language models as if they were informants. Surfaces the corpus lens — how a model organizes everyday vocabulary, refracted through training and alignment. Open data, reproducible, model-to-model.