> It's hard to describe it as a feature, but I'm afraid it
> falls into the category of implementation artifact. What did
> you expect to get back as the db_key of a record in each of
> the trigger types?

Exactly what you showed in your example! Amazing insight, you have! :-)

Triggers fire on all events(Ins,Upd,Del), and record the changes into a
central 'delta' table.
The delta table is inspected on a regular basis, and Delta's are then
migrated to the 'Replication Cortex', and redistributed cluster-wide.
That table contains the source table, field, row, old value, new value, and
a timestamp.

If rdb$db_key isn't predictable, I'll have to have a rethink...

Table design:
CREATE TABLE mmk$delta (
table CHAR(31) NOT NULL,
field CHAR(31) NOT NULL,
row char(8) character set octets NOT NULL,
oldv BLOB sub_type text,
newv BLOB sub_type text,
stamp TIMESTAMP default 'now' NOT NULL,
opcode INTEGER NOT NULL, /*1:ins,2:upd,3:del*/
PRIMARY KEY(table,field,row,stamp,opcode)
);

Nige.