>
> The point is you can't resolve a name like "schema_1.something" unless
> you have USAGE on schema_1. So the RI-checking query, which is run as
> the owner of the table, fails at parse time.
That certainly makes sense for user_2 that owns the reference table and is blocked by not having usage on the reference table’s schema.
But, user_1 owns both schemas and has usage on both but no longer owns the reference table in one schema. Why is user_1’s insert on the referencing table failing? Is the validation of the FK no longer done as user_1?