Task → do → verify
Build-along · ~10 min at your own keyboard
The single biggest difference between people who get magic from AI and people who get slop: the loop. Give it a task, let it do the task, give it a way to verify its own result. For a methods generator, verification means checking every parameter against the run record AND confirming no sentence states a result, before you ever see it. And one good run proves less than it seems. The model is non-deterministic: the same prompt on the same files can come back different next time. So the check has to run every time, not once when you first set it up.
Where these ideas come from: Non-determinism (AI Hero dictionary). Paraphrased here.
Your starter: Run brief with a definition of done
PromptMake it yours: Name your own figures or runs and the record files, change the two checks if your output is not legends or methods, and pick the one value you will remove for the broken case.
TASK Draft [legends / methods paragraphs] for [which figures or runs]. Inputs: [the run records or exports, by file name]. DEFINITION OF DONE An item is done only when both checks pass: 1. PARAMETERS TRACED: every number, unit and n appears in the run record. Quote the line it came from. 2. NO CLAIMS: no sentence states an outcome, a significance or an interpretation. VERIFY BEFORE YOU SHOW ME When finished, give a table: item | parameters traced (VERIFIED / FAILED) | no claims (VERIFIED / FAILED) | what is missing. If a value is not in the record, write [NEEDS: value] in the draft and mark the row FAILED. Do not fill it from a typical protocol. If you cannot check something from the inputs given, write CANNOT ASSESS and say what you would need. BROKEN CASE (for the test run) I will remove [one value, e.g. the n] from one record. That row must come back FAILED.
The coach checks that what you paste back is your own work. The starter pasted back unchanged won't pass.
Do this
0/4Share what happened
Paste your definition of done, the verification output, and proof the broken case got caught.