The methods section written from the script
Lesson 5 of 6 · From chat to script
1 · Learn the move · Verified-data override
The methods paragraph a reviewer can check is the one derived from the script - because the script is what ran. Written from memory, methods sections drift: the threshold you remember is the one you meant to use in April, not the one in the code. The override discipline: paste the script and have the model draft methods ONLY from what the code shows. Every threshold, test, and exclusion must point to a line; anything the code does not show - versions, a manual step, the reason behind a cutoff - becomes [NEEDS: ...] instead of smooth prose. Software versions come from your environment file, not memory, and any proposed citation enters UNVERIFIED until you check it - methods sections are where plausible fabricated references hide. The draft is a draft: you supply the why, and you sign it.
Draft a methods paragraph from this script ONLY - [paste script]. Rules: every number, test, and exclusion in your text must exist in the code; cite the line or function next to each claim in a bracketed note I will strip later. Anything methods sections normally state that this script does not show - software versions, the rationale for thresholds, data provenance - write as [NEEDS: ...], never from memory or convention. Software citations you propose are tagged UNVERIFIED until I confirm them against the actual papers. Do not smooth, do not fill, do not conclude anything about results.
2 · Your turn. You write the prompt
The manuscript deadline is Friday, the analysis lives in one 300-line script plus an environment.yml, and your co-author's methods draft says "standard quality filters were applied" - which is prose for "nobody checked." Generate the paragraph from the code itself, gaps marked.
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.