Research
Run a reproducible analysis
Code, figures and a record anyone can rerun to the same numbers.
Fill it in
What to analyse.
What you are testing.
Seeds, versions and the data checksum.
Your prompt
Run the analysis. Question: [the question] Write it as a script that runs end to end from the raw data. A notebook executed out of order is the standard way a result becomes irreproducible, and the author is usually the one who cannot reproduce it six months later. Pin everything: the random seed, the library versions, and a checksum of the input. Without those, "reproducible" means it ran on your machine today. State the analysis plan before running it, and say plainly if you changed it after seeing the data. Produce the figures from the same script that produces the numbers, so they cannot disagree. Report the effect size and the interval, not only whether it passed a threshold.
Use Run a reproducible analysisOpens with everything above already filled in.
Why this works
A notebook executed out of order is the standard way a result becomes irreproducible, usually for its own author six months later. This is a script that runs end to end, with the seed, the versions and a data checksum pinned.