odmlib

Use the odmlib define_2_1 model classes to build the document as objects, assigning Study and MetaDataVersion as single objects rather than lists. Then call validate() with an OID checker and write_xml() to serialize, optionally schema-checking with the bundled Define-XML 2.1 XSD.

Call validate(collect_errors=True, oid_checker=create_oid_checker(...), conformance_checker=MetadataSchema()) on the loaded odmlib object to enumerate element-order, OID integrity, and conformance errors in one pass. For full schema validation, use ODMSchemaValidator with the bundled ODM, Define-XML, or ARM XSDs.

Calling ET.tostring(obj.to_xml()) produces markup with no namespace declarations, so ODM re-parses into no namespace and every Study is silently dropped while FileOID still reads back. Use to_xml_string() or write_xml() instead, which attach the required xmlns declarations.

In odm_1_3_2 and odm_2_0, Study and MetaDataVersion are lists accessed as odm.Study[0].MetaDataVersion[0]. In define_2_1 and arm_1_0 they are single objects accessed as define.Study.MetaDataVersion, and indexing them raises an error.

Yes, wrap the load in permissive mode via open_odm(path, permissive=True) or the permissive() context manager to relax required-attribute, value-set, type, and format checks. Permissive mode is intended for inspection and repair; fix the issues, return to strict mode, and validate before writing output.

Load the Define-XML document with the define_2_1 loader, then call DefineFlattener(odm).flatten_all() to get named DatasetJSON datasets for study, datasets, variables, codelists, and more. Write each with write_ndjson(), since NDJSON is the Dataset-JSON v1.1 wire format.