Reproducible R is more than set.seed()

Record what actually ran

Keep the code revision, R version, package library and immutable input identifiers with each analytical run. A lockfile helps record package versions, but the operating system, external libraries and remote data sources can still affect a result.

Make dependencies explicit

A script that works only after several manual console commands is difficult to review. Start from a clean session, parameterise the data cut and separate source inputs from derived outputs. Use a pipeline tool when it improves dependency tracking, rather than adding one simply to increase the technology count.

Test the reconstruction

Ask someone to restore the environment and regenerate selected outputs using the documentation alone. Compare numerical results with justified tolerances and investigate unexpected differences. Keep a record of what cannot be reproduced exactly, including services whose underlying models may change.

Sources & further reading

• CRAN — renv
• CRAN — targets

Sources checked 20 September 2026. Implementation commentary reflects a practical review perspective; source material may change.