Why a fluent answer is not a checked answer
Lesson 1 of 6 · Run your controls
1 · Learn the move · Cognitive verifier
The same machinery that writes a clean sentence writes a wrong one, at the same speed, in the same tone. Fluency is not evidence. You already know this from the bench: a chromatogram can look beautiful and be garbage if the blank was skipped. Treat every model output as a draft carrying claims, and make the model itself sort those claims by support: supplied by you, textbook-level general knowledge, or generated just now in this answer. That third pile is where fabrication lives. The sort costs one follow-up prompt and turns a wall of confident prose into a short list of things to check. The model drafts and checks. It does not conclude.
Answer this question, then decompose your answer into numbered claims. Label each claim GIVEN (from material I supplied), GENERAL (textbook-level knowledge), or GENERATED (produced in this answer, no source). Flag every GENERATED claim as UNVERIFIED. Do not state a conclusion about my specific case. Question: [question + any data you are supplying]
2 · Your turn. You write the prompt
You asked a model why your qPCR standard curve shows an efficiency of 112 percent, and the answer reads like a textbook page - organized, confident, plausible. You cannot tell which parts are established amplification chemistry and which parts it invented to fill the paragraph. Write a prompt that sorts the claims before you touch your assay.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.