Move 3: anything done twice becomes a script
Lesson 4 of 6 · What the model can and cannot see
1 · Learn the move · Refinement loop
The first time is exploration. The second time is a pattern, and a pattern that lives in a chat window is a liability: a chat transcript is not a method section, and six months from now it is not even a memory. Move 3: anything done twice becomes a script you keep, validate, and re-run. The refinement loop is how a chat becomes that script honestly. Draft: paste the working exchange and have the model extract the steps into a script. Critique: have it attack its own draft against the transcript - what did the chat do that the script does not, what does the script assume that the chat checked by eye? Validate: run the script on the input from the original session and require identical output before it ever touches new data. The loop matters because transcription is lossy - your eyeballing of an intermediate result was a real step, and the script needs it as an assertion, not a vibe.
Below is a chat transcript where we [task] successfully, step by step. Convert it to a script in three passes. PASS 1 DRAFT: extract every operation into a script, in order. PASS 2 CRITIQUE: compare the script against the transcript - list every check I did by eye that the script does not do, and every assumption the script makes that the chat verified interactively; convert each into an assertion or a [NEEDS: decision]. PASS 3 REVISE: final script plus a validation instruction - run on [original input], expected output [original output], any difference is a failure. Transcript: [paste]
2 · Your turn. You write the prompt
Twice this month you have walked a model through the same cleanup of your plate reader kinetics export - find the header, drop the instrument footer, join the plate layout, flag saturated wells. The second time took as long as the first, and next month it will happen again. Write the prompt that turns this month's transcript into next month's script.
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.