Stop 1 of 7 · Read once
The line
The one rule that makes every other stop safe, and the study Project that carries it for you.
One sentence, first
This course is not a clinical decision tool. Nothing in it helps you decide what to do for the patient in front of you, and no prompt in it asks Claude anything about a real person. Rana's judgement, her protocols and her medical direction are unchanged. What changes is the work off the floor: the credential she is studying for, the guideline nobody has time to read, the handout the educator asked for, the orientee who needs a competency signed, the committee that wants a draft protocol by the 20th.
What stays out
| Stays out of every chat | Comes in |
|---|---|
| Anything about a patient, including a de-identified version | Published guidelines and standards you downloaded yourself |
| Vent screens, waveform photos, films, the census board | Your employer's approved patient education content |
| A colleague's performance details with their name attached | Exam content outlines and your own study notes |
| Anything inside an incident review or a legal hold | Your resume, your CE certificates, your own drafts |
De-identified is where people talk themselves into trouble. A rare presentation, on your unit, last Tuesday, is one person with the name taken off. And a rule with an exception is a rule you have to weigh at 0300 at the end of a twelve, which is precisely when you will get it wrong. One rule, no exception, nothing to decide.
The second rule is quieter and does as much work. Every clinical sentence Claude writes for Rana names the document, the version year and the page, and Rana opens that page. Not "per the guideline". A model can write a confident recommendation no society ever made, in the register of one that did. A page number makes that cheap to catch.
The study Project
Everything in this course runs inside one Project so the rules are never retyped. Files: the exam content outline, her own notes, the approved education pack, her CE log.
You help me, Rana, a respiratory therapist in critical care with six years on the floor. I use you for study, teaching, writing and paperwork. I never use you for care decisions and you never offer one. Rule one. I will never give you information about a patient. If anything I paste looks like patient information, even with the name removed, stop and say so instead of answering it. Rule two. For anything clinical, work only from the documents in this Project or a file I attach to the chat. Do not answer a clinical question from memory. Put the document name, the version year and the page or section in brackets after every sentence that makes a claim. If the documents do not cover it, write that it is not in the sources I gave you, and stop there. Rule three. Never raise the strength of a recommendation. If a source calls something conditional or low certainty, say conditional or low certainty in the same sentence. Voice: plain and short. No exclamation marks, no emoji, no warm-up paragraph. Spell out any abbreviation the first time it appears in anything for patients or families. When I ask you to quiz me, ask one question, wait for my answer, then explain.
Here is a prompt I was about to send you: [paste your draft prompt]. Do not answer it. Tell me whether any part of it contains information about a patient, including anything de-identified, and then rewrite it to ask the same underlying question using only published sources and my own study notes.
The exam content outline in this Project is for the credential I am studying for. List the sections it covers and, for each, the kind of document I should add here so answers trace to a source rather than to your memory. Do not write any clinical content yet.