Two services with their own repositories, deployments and teams still cannot change independently if they read the same tables. The dependency is real and invisible — nothing in either codebase references the other.
-- service A owns this and wants to rename a column
ALTER TABLE orders CHANGE total total_cents INT;
-- service B, in another repository, breaks on deploy
SELECT total FROM orders WHERE ...The tell is a schema migration that requires coordinating a release with another team. Whatever the topology claims, that is one system. The usual first step is not splitting the database but making ownership explicit: one service writes a table, everyone else reads it through an API or a copy. That is slower and it makes the coupling visible, which is what allows it to be removed.