Unpublished data, and why disclosure is not undoable
Lesson 1 of 6 · What you cannot paste
1 · Learn the move · Constraints + negative instructions
Every other mistake in this course is recoverable. Disclosure is not: once unpublished data leaves your control, you cannot recall it, and what it cost you gets decided later, by other people - a journal, an examiner, a sponsor. So the paste decision is never a convenience decision. This area will never tell you something is fine to paste, and neither will any tool - the people who can answer that are your TTO, your sponsor agreement, and your IRB, and every lesson here routes to them. What you can do today is write prompts that carry the constraint: the model works from a described shape of the data, placeholders instead of values, and is explicitly forbidden to ask for the real thing.
Help me draft [document] about an unpublished experiment. Constraints: do not ask me to paste the data. Work only from this description: [design, sample sizes, what was measured, direction of effect - no values]. Use bracketed placeholders like [VALUE-1] wherever a number belongs, and keep a placeholder key at the end so I can fill them in locally. If any part of the task cannot be done without the real values, say which part and stop there.
2 · Your turn. You write the prompt
You need a first draft of an abstract paragraph describing results that are unpublished and, if the project holds, patentable. The values cannot leave your machine, but the writing help would genuinely save the evening. Write a prompt that gets the drafting without the data.
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.