What QC means when the model cannot see the sample
Lesson 1 of 6 · Cleaning and QC
1 · Learn the move · RTO. Role · Task · Output
The model never saw the sample, never smelled the media, never watched well D7 get pipetted at 6 pm on a Friday. QC through a model is QC over the slice: distributions, counts, control behavior, structural consistency - the shadows your experiment casts into the numbers. That is genuinely useful, and it is strictly less than QC. So the first prompt in any QC session sets the role honestly: a QC drafter that enumerates what it can check from the supplied data, what it cannot assess at all, and never blurs the line. The output format carries the honesty: a checklist with a CANNOT-ASSESS column that stays populated. A model that says your data passed QC has overclaimed by exactly the width of everything in that column - the model drafts checks and runs them on numbers; passing your experiment is not in its gift.
You are a QC drafter working only from supplied data - you have never seen the samples or the run. My experiment: [design, what was measured, what data I will supply]. Output two lists before any checking: CAN CHECK - every QC dimension assessable from the data I described, each with the specific test you would run; CANNOT ASSESS - every QC dimension that lives in the lab, not the numbers (sample integrity, handling, timing, instrument state), stated plainly. Then run the CAN CHECK list on: [data]. Never present the result as the experiment passing QC.
2 · Your turn. You write the prompt
Your new student ran her first ELISA plate and asked the lab's chat tool if her data was OK; it said the data looks good, and she wrote that in her notebook. You need her next prompt to teach her - and the model - where checkable ends and unknowable begins. Write that prompt.
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.