The brief you steered: written for the board investment committee, backing the Q4 procurement budget, with the 2026 forecast included only if it comes with sources.
Fluent prose hides which sentences are load-bearing. Ground marks every claim, ties the ones it can to a source, and is explicit about the ones it cannot. Click any marked claim to see what stands behind it.
Four claims. Same words the model would have handed you anywhere else, with one difference: each claim is a control, and its status is on its face.
Select a claim in the draft above.
Two numbers, because they move for different reasons. Only new evidence raises the first. Rewording a claim can only lower the second.
Claims a reader could check for themselves.
Assertions the sources do not carry. This has to reach zero before the brief can go out.
What was checked, what was changed, and what is still missing.
The forecast is honest now, but still untraced. The only source that would fix it is live supplier pricing, which sits in three vendor systems. That is a job for an agent, and it touches real money.
Go to Oversee →Press play to watch this run start to finish.
Four claims, written in the same confident voice. Two are traceable, one is contested, and one has nothing behind it at all. Nothing in the prose tells you which is which, so the interface has to.
About 55 seconds, narrated. Synthetic data. Pause or take over at any point.