Caption: 
A new Southwest Research Institute-led study compared AI-generated lunar crater catalogs, discovering that their published performance metrics drop sharply when the databases are compared to human-built resources. Scientists use crater catalogs to estimate surface ages, reconstruct geologic histories and study how planetary surfaces evolve. This example maps the results from the different catalogs, including the human-derived data in green, showing inaccuracies in size and location associated with automated systems — flaws that could skew scientific studies.