Why do my ATP instances sometimes go from "provisioning" to "unavailable"?

Summary:

We've been using the OCI DatabaseClient Python API to call create_autonomous_database. This works very well, but occasionally a newly created ATP's lifecycle_state will go from "PROVISIONING" to "UNAVAILABLE" and never recover

Content (required):

After calling the OCI python function create_autonomous_database, a new ATP instance is created and it's lifecycle_status starts as "PROVISIONING". After some time, normally it will then change to "AVAILABLE" and be ready for connections. To be clear, we are creating clones using create_autonomous_database's SOURCE_BACKUP_FROM_ID and providing the backup OCID, which is usually the 2nd to last known good backup of our source ATP OCI db.