Would be interesting to hook up a much simpler LLM as fact checker to see when errors are introduced.

If I had to place a hidden target it'd probably be around RNGs or publicly exposed services..