Author's checklist
This checklist implements "A companion repository holds six parts". It is for the person who ran the analysis. Do the steps in order. Institutional policy and data use agreements come before every step.
Code
-
Put the scripts, notebooks or spreadsheets that produced each table and figure in one repository.
-
Keep spreadsheet formulas in place. Do not paste values over them.
-
Remove credentials, keys and personal data from the files and from the repository history.
Environment
-
Record the version of each package and program the code uses, in a lockfile or a container definition.
-
Run the code once in a clean environment made only from that record.
Data reference
-
Write the DOI or accession of the exact data snapshot the paper cites.
-
If the data is patient-level, keep it in its governed repository. Put a synthetic or aggregate copy with the same structure in the companion repository, so the code runs end to end.
-
If the data has no patient-level content and its terms allow it, put the snapshot in the companion repository.
One entry point
-
Write one command, or one numbered list of steps, that goes from the data to every result.
-
Ask a colleague who did not do the analysis to follow it in a clean environment. Fix each step the colleague had to guess.
Declared outputs
-
List each table and figure in the paper.
-
For each, write the file that the entry point writes for it.
Metadata and links
-
Add a citation file and a software metadata file to the repository.
-
State a licence.
-
Archive the release that the paper uses, so that it gets a version DOI.
-
Put the version DOI in the paper. Put the paper’s DOI and the data record’s DOI in the repository.
After publication
-
When you rerun the method on new data, make a new release. Do not change the release that the paper cites.