The Croissant metadata format supports describing relationships between data resources (e.g., using references). Currently, the croissant-maker appears to treat each file independently. It would be highly beneficial to automatically detect potential relationships between tables based on column names and data content (e.g., matching a subject_id in a CSV file with a subject_id in a Parquet file). This would significantly enrich the generated metadata and make the resulting Croissant dataset more useful for automated tools.