How to safely revert a DynamoDB Global Table version upgrade after a failure?

Upgrading DynamoDB Global Tables to version 2019.11.09 introduces significant changes to how replication and write conflicts are handled. When an upgrade process is initiated across multiple regions, a failure during the transition can leave the table in an inconsistent state where some regions are on the new version and others remain on the legacy version.

The primary constraint is maintaining data integrity and availability without incurring significant downtime or permanent data loss during the recovery process. There is uncertainty regarding whether a direct downgrade is supported once the upgrade has been partially committed.

What is the recommended procedure to recover a Global Table from a failed version upgrade? How can the table be returned to a consistent version across all regions? What methods exist to verify that no data was lost during the rollback?