Evidence kit
A workflow intake sheet, an AI-use register, a one-page context-of-use template and a reviewer checklist, ready to adapt to your SOPs.
Early-stage project, open to a few teams
Regulated analysis your team can trace, check and replay.
Reprosia helps pharma and biotech statistical teams make their analysis work reviewable: consultancy, validation and migration, backed by open-source tools we maintain, with a growing focus on AI-assisted analysis.
What we do
ConsultancyValidationMigrationTrainingOpen-source tools
Audit ledger, live in your browser
Building chain…
Each record stores the hash of the one before it. Press Tamper on any entry and verification fails. This demo runs SHA-256 in your browser; it is not output from regulog.
What Reprosia does.
One consultancy, one method: make the analysis reviewable. Open-source tools are one part of it, not the whole offer.
Projects
Consultancy projects
Regulated R implementation, IQ/OQ/PQ qualification, reproducibility audits and SAS-to-R migration, scoped and quoted up front.
See services Contract and advisoryEmbedded and on call
Statistical programming support, second opinions and retainers, plus training in R and reproducible workflows.
Ways to work together Focus areaAI-use governance
What to log when AI helps with an analysis, who reviews it and how to replay it, and how to measure an AI coding agent before relying on it. Starts with a free evidence kit.
See the approach Open sourceTooling
R packages on CRAN for audit logging, provenance and reproducibility. Usable without Reprosia.
See the toolingThree ways to start.
Begin with something you can use this week, then go deeper only if it helps.
Intro call
Talk through where AI enters your analysis. You leave knowing which records a reviewer would ask for and which you probably have already.
Evidence sprint
About two weeks on one workflow: a gap report against the ten FDA and EMA principles, an audit log running on one real script, and a replay check. Scoped and quoted up front.
Try it before you talk to anyone.
Working tools that run entirely in your browser. No model is called and nothing is uploaded.
Interactive
Audited AI assist simulator
Scope a task, watch a simulated draft stream in, record reviewer decisions and replay the run. Every step joins a hash-chained log you can verify, tamper with and export.
Open the workbench InteractiveEvidence readiness check
Ten questions, one per FDA and EMA principle. You get a radar of where your AI-assisted workflow stands, a ranked gap list and a report to download.
Run the checkAsk the evidence map
Search plain answers on tamper detection, Part 11, review records and what is out of scope.
Ask a questionPrinciples, mapped
The ten joint principles, matched to what exists today and what does not.
See the mapHow it fits your stack
An evidence layer between your R work and your quality system.
See the platformOpen-source tooling behind the work.
Open-source R packages on CRAN. Engagements use them where they help and never depend on them.
| Package | What it does | Status |
|---|---|---|
| regulog | Tamper-evident, hash-chained audit logging designed around 21 CFR Part 11 and EU Annex 11. Ships with IQ/OQ/PQ qualification scripts. | On CRAN |
| lineager | Row-level data provenance and exclusion tracking, with provenance reports aligned to the CDISC Reviewer's Guide. | On CRAN |
| reproducr | Reproducibility auditing: session capture, dependency locking, script certification and drift detection. | On CRAN |
| bayprior | Bayesian prior elicitation for clinical trials, with conflict diagnostics, sensitivity analysis and regulatory reporting. A statistical methods tool that sits beside the audit tools rather than inside them. | On CRAN Methods |
Regulators are signalling what they will want to see.
In January 2026 the FDA and EMA published joint guiding principles for AI in drug development. They are non-binding. Our reading is that they lean towards documented data provenance, controlled and reproducible environments, and records of human review, though the wording is high level and open to interpretation.
The approach page goes through the ten principles one by one and says how far Reprosia gets you on each, including where it does not help. For a concrete example, see the worked example, a real run of the tools with its output.
Our reading draws on a third-party summary. Check the official text before you rely on any wording.
Preparing R work for regulatory review?
Describe what your QA group or a reviewer would need to see, whether that is validation, a migration or the use of an AI assistant. Engagements are taken on in small numbers.