The agent said 1 row. The database changed 301. Caught by Deedglass.
The task: “Mark order 42 as shipped.” The agent is scripted; the database result is a real run on the /play store.
The run
ClaimedWhat the agent says
UPDATE orders SET status = 'shipped' WHERE id = 42
“Done, 1 row.”
1row
ProvenWhat the database committed
public.orders, 500 rows. Order 42 is outlined.
301rows
- 301 rows changed in orders (status)
- It ran. The tool reported 1 row.
Doesn’t match
The agent said 1 row. The database changed 301 rows.
The hidden catch: A hidden trigger. A legacy trigger ships up to 300 other open orders whenever one order ships.
In our benchmark, the same kind of case (“A trigger changes 300 more rows than the statement names”): Deedglass caught after it ran.
Share it.
