Laws such as California's Training Data Transparency Act and the EU AI Act yield broad dataset summaries that fall short of meaningful accountability. The article argues that disclosures should also explain how training data is acquired, filtered and evaluated, including the benchmarks and judgment calls involved.