This seems like the right approach to me.

An approach I've taken is temporal tables w/ Application and UpdatedBy fields. That gives you a permanent record of every change, what application did it, and what user performed the action, at what time, and then lets you query the database as if you were querying it at that point in time. You can add triggers to fail CRUD if those fields are not updated if you want to get really paranoid.

There's a lot of overhead to this in terms of storage, so it's not suitable for high-throughput or cost-constrained transactional systems, but it's something for the toolbox.

Even if it doesn't you can start by adding the application name to the connection string and you could probably do something gnarly with triggers to write that in a table and get it pushed in the logical replication.