The 60-second check before you use any output
Lesson 6 of 6 · Run your controls
1 · Learn the move · Recipe pattern
Every check in this area compresses into one fixed recipe, run on every output you are about to use, in order. One: controls - did this run carry a known, and did the known pass? No control, no confidence. Two: numbers - is every number traceable to something you supplied or a source you opened? Three: references - does every citation carry a state, and is any still UNVERIFIED? Four: conclusion smell - does any sentence claim more than the data in front of it? Drafts and checks are usable; conclusions are yours to make. Five: disagreement - does anything contradict something you already know? A contradiction is a stop, not a footnote. Sixty seconds, because it is the same five questions every time, and the recipe never gets skipped on a busy day - busy days are what it is for.
Run the pre-use check on the output below as a checklist, one line per step, PASS / FAIL / N/A with a one-sentence reason: (1) CONTROLS - was a known-answer item included and did it pass; (2) NUMBERS - every number traceable to supplied material; (3) REFERENCES - every citation carries VERIFIED / UNVERIFIED / FAILED, none UNVERIFIED; (4) OVERREACH - any sentence concluding beyond the supplied data, quote it; (5) CONFLICT - anything contradicting the known facts I list here: [known facts]. End with USE / DO NOT USE YET and what would clear it. Output to check: [paste]
2 · Your turn. You write the prompt
It is 5:40 pm, your lab meeting is at 6, and a model just produced the summary slide text for your subgroup's data. It looks fine. Everything looks fine at 5:40. Write a prompt that runs the full 60-second check so that what goes on the slide has been through the same gate as everything else you present.
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.