What 'Reproducible' Actually Requires

A visiting researcher asks to reproduce your nitrate trend analysis from Mission 05. You hand over a folder: a dozen scripts with names like `final2.py` and `analysis_REAL.py`, no record of which version produced the report's numbers, and a dataset you can no longer exactly recreate because you've since 'cleaned it up a bit more.' Reproducing your own six-month-old work turns out to be nearly impossible, for you, let alone them.

Reproducibility (the ability for someone else, or future you, to redo your exact work and get the exact same answer) means someone else, or future you, could take your raw data, your documented steps, and your code, and arrive at the same result. It requires more than 'the code works': it requires a clear, unambiguous path from raw data to final figure that doesn't depend on memory, verbal explanation, or files that no longer exist in the state they were used.

It is like the difference between a recipe that says 'add a splash of milk, cook until it feels right' and one that says '120ml of milk, simmer for 8 minutes.' Both might make a good dish once. Only the second one can be handed to a stranger and reliably repeated.

Large-scale audits across multiple scientific fields over the past decade found that a majority of published computational results could not be reproduced by independent teams using the original authors' own described methods, a finding significant enough to be called a 'reproducibility crisis.' The practices in this mission exist specifically to prevent your work from becoming another data point in that statistic.

Explain to a new researcher why 'the code runs' and 'this is reproducible' are different claims.