Split brain part:
In authoritative shards, the router cannot have a more up-to-date DB version than the primary. In the rare case of the replica set suffering a split brain, a node might think it is a primary but doesn't really have the quorum for it. In that case, DDLs could have advanced the database version, and routers could have gone further ahead. It is rare, but we need to handle that case for the primary CRUD path as well.

Term change handling:
For authoritative shards, a secondary will wait for oplog replication to catch up with the timestamp in the DB version. We need to think carefully of any replication state changes that could happen during that time and how to deal with them:

• The secondary itself might go through an election - and either become the primary, or stay the secondary.
• Other nodes might go through an election and a term change might happen

Other than the waiting secondary itself becoming a primary, intuitively it feels any other transition should be okay - let's think more in this ticket and commit any changes.