Reproducible Environments (2026)

Environment and package management for regulatory submissions - locked-down vs. open environments, cross-OS reproducibility, and tooling like renv, rv, and uv

2026
SCE
Published

September 13, 2026

Reproducible Environments

See also: 2023 Discussion, 2024 Discussion, 2025 Discussion

The Submission Reproducibility Challenge

  • The “Recipe” We Ship: For submissions, sponsors must record the environment - R version, package versions, proprietary packages - so regulators can attempt to reproduce results
  • OS Mismatch: Sponsors typically develop on Linux (often containerized), while regulators run Windows - the recipe must be detailed enough to bridge that gap and get regulators as close as possible to the original environment
  • How Far Do We Go?: How much testing is required before shipping is dominating discussions

Locked-Down vs. Open Environments

  • Both Are Needed: Organizations need a spectrum of options for different use cases, not a single environment strategy
  • Locked-Down Advantages: Reproducibility is baked in - grab the container, revisit later, and it just works
  • Open Environment Challenges: Reproducibility problems happen when users hit package/environment management issues for the first time and only learn about them after something breaks
  • Managing Both: Strategies exist for managing open environments, but there are still real bumps

Tooling: renv, rv, uv

  • Learning Curve: renv, rv, uv and similar tools require users to understand why they exist and what they’re doing behind the scenes - a tall order for the average user
  • Feedback to Vendors: A direct request to the community from Posit - when renv (or similar) doesn’t work, file it. Capture what you were trying to do, include log files, and send a GitHub issue (or reach out via your rep if it must be private). Without that context, the problems are hard to fix

Base Starting Point

  • Every Project a Snowflake?: If every project starts from base R and layers on renv/rv from scratch, you’re rebuilding from square one each time
  • A Common Core: Better model - agree on a shared core set of packages/elements consistent across projects, then layer only the project-specific extras on top
  • Reduces Variability: A common baseline keeps the surface area small and makes environments more predictable across teams

Overall Takeaway

Reproducible environments for submissions require sponsors to think in terms of “recipes” that survive an OS change on the regulator’s end. Organizations need both locked-down and open environments, education on the tooling powering them, and a common core to avoid every project being bespoke.