THE CONFIDENT LIE
Which claims survive contact with the source?
After this station you can take fluent, confident machine output and tell, claim by claim, which parts survive contact with the source — and which were invented.
The most dangerous machine output is not obviously false. It is plausible, smooth, confident — the kind of thing a smart person might say. And it may be wrong.
The skill is not to admire the writing. It is to attack it: line by line, claim by claim, against what the source actually says.
You extract each claim, find its source, and compare what the source says to what the machine said. Then you grade it.
Your G3 sources, a table, and a second AI used as an adversary — not an authority.
Take an AI summary of your topic and break it — then leave only what holds.
Your verification holds when:
The lie did not arrive screaming. It arrived formatted. It arrived helpful. It arrived with bullet points.
That is why you check.
Not because you distrust intelligence. Because you respect evidence.
You can now verify the machine. Next — MASCHINENRAUM: step back and inspect the whole workflow you have built.
The most dangerous lie is the one that sounds like work already done.