The point I was trying to make is that the alternative—writing all your DB activity into the DB—means that you're already, 100% guaranteed, creating more overhead than the heaviest possible journal-safety properties would create. You're writing unabridged journal entries into the DB, where they then get even more journal attached.
There is no world where it makes more sense to write a copy of everything you do into an "everything I did" table, than it does to use an append-only / immutable / log-centric DBMS (where "Postgres with an ETL pipeline pointed at its WAL logs" is a rather hacky immutable DBMS.)
If you know, before you ever start coding, that you will need to know everything that ever happened, writing everything twice—when your DBMS is already writing everything twice—is a rather silly way to choose to go about solving your problems. You're already in a world where you need different guarantees than a traditional (ephemeral) RDBMS gives you; you just haven't realized it yet.
