Test an agent with an answer key.
Eight hidden events. Seven known causes. One that must stay unsolved.
How do you know a root-cause agent works? Give it problems you already solved.
BottleCo runs on sample data with eight events hidden inside: a late preform delivery, a drifting capper, rejects after a changeover, and more. I know the cause of each one.
The test is simple. Seven events must resolve to their true cause. The eighth has no clear cause in the data. The agent passes only if it says so.
That last test matters most. An agent that always finds a cause is not smart. It is guessing with confidence.
See it in BottleCo. Open BottleCo