FDA says yes to R, no to your Docker image

The R Consortium Submissions Working Group’s May 2026 post on Pilot 4 carries the first specific FDA verdict on container-based Shiny submissions, and it is not the one most teams were planning around. Reviewers ran both the WebAssembly bundle (submitted via eCTD on 20 September 2024) and the combined WebAssembly + Docker resubmission (June 2025), and concluded that Docker, as currently implemented, does not meet the robustness standard required for review. The stated reasons are unromantic: Docker Desktop requires WSL, which may be restricted on FDA-issued laptops, and the agency does not support Docker Desktop without a paid licence or Docker Community Edition. Build performance was inconsistent on top of that.

The counterintuitive finding is that reviewers preferred WebAssembly — the option the working group expected to be the harder sell — because it runs in a browser. The asterisk is that Wasm still required renv for dependency restoration and a local web server to render the app, so the reproducibility burden on the sponsor is unchanged. “Browser-only” turned out to mean “no admin rights needed,” not “no environment to reproduce.” For programming leads, that reframes the packaging question: the win is reviewer-side accessibility, not sponsor-side simplification.

What is actually working in 2026

The “what works today” counterweight is Johnson & Johnson’s Hybrid R/SAS case study: SAS for ADaM derivation, R for every Table, Figure, and Listing. One submission approved within the planned timeline, a second in progress with positive agency feedback. The reproducibility trick is worth stealing — rather than transport a Linux-built renv.lock to a Windows reviewer machine, J&J ship an ADRG appendix (modelled on the Pilot 3 template) that walks the reviewer through CRAN/GitHub installation and generates the lockfile on the reviewer’s box, post-install. It is not elegant. It works.

Tucked into the same post is a quiet but operationally consequential item: the FDA Study Data Technical Conformance Guide, August 2025 version, now explicitly permits R-native file extensions in eCTD. The .txt-rename ritual that programmers have performed for years to keep eCTD validators happy is finally retired. Edit-check libraries, submission checklists, and your define-XML packaging scripts all need a line changed.

The blocker that will define the next 12 months

Pilot 5 tried to replace every .xpt with Dataset-JSON v1.1 across SDTM and ADaM. FDA accepted the structure, asked for rework on intermediate dataset handling, and the resubmission landed in January 2026. The catch is downstream: Dataset-JSON v1.1 is not yet supported by Pinnacle 21, so the pilot fell back to .xpt for validation. That is the operational shape of the next year — sponsors can submit Dataset-JSON, but cannot yet validate it with the tool the industry treats as authoritative, which means dual-format pipelines, doubled storage, and reconciliation logic nobody wants to own. The 2026 roadmap adds Pilot 6 (scaling ADaM and display programs beyond the toy Pilots 1–3 sets) and Pilot 7 (new simulated data, finally moving off the original CDISC pilot dataset).

Read across the three pieces, the throughline is consistent. Regulators are saying yes to R — formally, in the Conformance Guide; operationally, in named-sponsor approvals; technically, in the Pilot 4 review letter. The rate limiter has moved to the tooling on both sides: FDA laptop policy on the reviewer side, Pinnacle 21 release timing and cross-OS environment plumbing on the sponsor side. None of this is a regulatory earthquake. It is a long sequence of small, expensive integration problems, and they are now your integration problems.

Protocol read: The R submission question has shifted from “will FDA accept it” to “can both sides actually open the package” — and the answer depends on Pinnacle 21’s release calendar and your reviewer’s WSL permissions, not on methodology.

What to do now:

• Drop the .txt rename step from eCTD packaging SOPs and update define-XML / submission checklists to the August 2025 Conformance Guide.
• Defer Docker-based Shiny submissions; pilot WebAssembly delivery with a Pilot 3–style ADRG appendix for renv restoration.
• Track Pinnacle 21’s Dataset-JSON v1.1 support before committing any study to a .xpt-free pipeline; assume dual-format through 2026.