We update the table schema as we run into new fields, up to a limit. We also store the unstructured part of the data in a column that can be queried via json_extract_path_text.