The reproducibility badge reflects the status of the gallery repos. See those repos for a live demonstration of
repro_badge()in a real analysis workflow.
Know your R analysis will produce the same results tomorrow as it does today.
The problem
You finish an analysis. The code runs. The numbers look right. But are they stable?
Package updates change function behaviour silently. Stochastic code without a fixed seed produces different results on every run. A model fitted on one platform may return subtly different coefficients on another. Results that were correct in January may drift by March — with no error, no warning, and no obvious cause.
reproducr makes these risks visible and trackable, before they reach a journal, a regulator, or a collaborator.
What it does
Scans your scripts for known breaking changes across popular CRAN packages, flags stochastic calls missing set.seed(), and identifies locale-sensitive operations that behave differently across systems.
Certifies your outputs by hashing key results — model coefficients, summary statistics, p-values — so any numerical drift is detected automatically on subsequent runs.
Generates audit reports in three styles: a plain summary, a ready-to-paste academic methods paragraph, and a structured QC document with sign-off fields for regulated workflows.
Works with your existing setup. If you use renv, reproducr reads your lockfile automatically. If you don’t, it uses your installed library. No configuration required.
Installation
# Development version from GitHub
install.packages("remotes")
remotes::install_github("repro-stats/reproducr")Gallery
Real-world pipelines demonstrating the full reproducr workflow across different domains, environments, and regulatory contexts. Each repository is a complete, independently runnable analysis with CI that audits on every push, certifies outputs, detects drift, and updates the badge automatically.
| Example | Domain | Audience | renv | Report style | Badge | Walkthrough |
|---|---|---|---|---|---|---|
| reproducr-ecology | Ecology / penguins | General R users | No | minimal |  |
DEMO.md |
| reproducr-clinical | Clinical trials / oncology | Biostatisticians, pharma | Yes | pharma |  |
DEMO.md |
| reproducr-rwe | Real world evidence | Epidemiologists, HEOR | Yes | academic |  |
DEMO.md |
| reproducr-cmc | CMC statistics | CMC statisticians, regulatory affairs | Yes | pharma |  |
DEMO.md |
Together they demonstrate that reproducr is environment-agnostic — it works with or without renv, across domains, and at any level of regulatory rigour. Each DEMO.md walks through the complete pipeline with real output at every step.
Quick start
library(reproducr)
# Step 1: Audit your script
report <- audit_script("analysis.R")
print(report)
#>
#> -- reproducr audit report [2026-05-30 14:32] --
#>
#> Files scanned: 1
#> Packages found: 4
#> Calls detected: 23
#> R version: 4.4.2
#> Platform: aarch64-apple-darwin20
#> Versions from: installed library
#>
#> Next step: risks <- risk_score(report)
# Step 2: Score for risk
risks <- risk_score(report)
print(risks)
#>
#> -- reproducr risk score --
#>
#> HIGH: 1
#> MEDIUM: 2
#> LOW: 1
#>
#> [HIGH] dplyr::summarise (line 14 in analysis.R)
#> Check : changelog
#> Details : In dplyr 1.1.0, summarise() changed its default
#> grouping behaviour ...
#> Reference: https://dplyr.tidyverse.org/news/index.html#dplyr-110
# Step 3: Certify your outputs as a baseline
model <- lm(mpg ~ wt, data = mtcars)
certify(
outputs = list(
coefs = coef(model),
r_squared = summary(model)$r.squared,
n_obs = nrow(mtcars)
),
tag = "submission-v1",
script = "analysis.R"
)
#> reproducr: certified 3 output(s) [2026-05-30] under tag 'submission-v1'
# Step 4: After any environment change or package upgrade, check for drift
check_drift(
outputs = list(
coefs = coef(model),
r_squared = summary(model)$r.squared,
n_obs = nrow(mtcars)
),
against = "submission-v1"
)
#>
#> -- reproducr drift check vs 'submission-v1' --
#>
#> Verdict : ALL OUTPUTS MATCH
#> OK : 3
#> Drifted : 0
# Step 5: Generate a report
repro_report(report, risks, format = "html", style = "pharma",
output_file = "qc_report.html")
# Step 6: Badge your README
repro_badge(report, risks, output = "README")Core functions
| Function | Tier | Purpose |
|---|---|---|
audit_script() |
1 | Parse a script and extract all pkg::fn calls with version info |
risk_score() |
1 | Check calls against the breaking-changes database |
certify() |
2 | Hash and store analytical outputs as a signed baseline |
check_drift() |
2 | Compare current outputs against a stored baseline |
list_certs() |
2 | Inspect all certifications in a project |
repro_report() |
3 | Render audit report (text / Markdown / HTML) |
repro_badge() |
3 | Generate a shields.io reproducibility badge |
check_db_staleness() |
— | Check database entries against current CRAN versions |
The three-tier workflow
Tier 1 — Scan & score Tier 2 — Baseline & drift Tier 3 — Report & export
───────────────────── ───────────────────────── ─────────────────────────
audit_script() certify() repro_report()
│ │ │
▼ ▼ ▼
risk_score() check_drift() repro_badge()Use Tier 1 alone for a quick scan, or build the full pipeline for regulated or peer-reviewed work.
The breaking-changes database
The heart of risk_score() is a curated database of known cases where a package update silently changed function behaviour — not errors, not deprecation warnings, just different results.
Current coverage:
| Package | Entries | Examples |
|---|---|---|
dplyr |
5 |
summarise() grouping change (v1.1.0), across() naming (v1.1.0) |
tidyr |
3 |
nest() interface rewrite (v1.0.0), pivot_wider() duplicate handling (v1.2.0) |
ggplot2 |
3 | Default colour scale change (v3.4.0), aes() scoping (v3.5.0) |
readr |
2 | vroom backend switch, column type guessing (v2.0.0) |
purrr |
2 | Error handling change (v1.0.0), map_df() deprecation |
stringr |
1 |
str_c() NA propagation (v1.5.0) |
lubridate |
2 | DST arithmetic (v1.9.0) |
broom |
1 | Column renaming (v0.8.0) |
data.table |
2 |
fread() type detection, melt() factor→character |
lme4 |
1 | Optimizer tolerance change |
caret |
1 |
train() interface changes (v7.0.0) |
rstan |
2 | Array syntax change (v2.26), extract() permutation default |
base R |
5 | RNG change (R 3.6.0), hclust() tie-breaking (R 4.0.0) |
The database is kept current via a weekly automated check — see check_db_staleness(). Community contributions are welcome via reproducr-db.
Risk checks
"changelog" — Breaking changes database
Checks every detected pkg::fn call against the built-in database. A call is flagged only if the installed version falls within a known risky version window (from_ver, to_ver].
Risk levels: - HIGH — output values can change silently with no error - MEDIUM — argument renamed or deprecated; may error or produce different output - LOW — minor behavioural note; output unlikely to differ in practice
"seed_check" — Missing set.seed()
Flags any call to a stochastic function (rnorm, sample, rbinom, etc.) where no set.seed() is found within 50 lines above the call.
"locale_check" — Locale-sensitive operations
Flags functions whose output depends on the system locale (sort(), format(), strftime(), etc.) — relevant when code runs on servers in different countries or with different OS locale settings.
Certification and drift detection
# After your analysis completes, certify the key outputs:
certify(
outputs = list(
model_coefs = coef(my_model),
final_n = nrow(results),
primary_pval = tidy(my_model)$p.value[2]
),
tag = "pre-review",
script = "main_analysis.R"
)
# Three months later, after any environment change:
check_drift(
outputs = list(
model_coefs = coef(my_model),
final_n = nrow(results),
primary_pval = tidy(my_model)$p.value[2]
),
against = "pre-review"
)Certifications accumulate in a .reproducr.rds file in your project root. Commit this file to version control — it is your audit trail.
Report styles
"academic"
A ready-to-paste methods paragraph for journal submissions:
All analyses were conducted in R (version 4.4.2) on macOS 26.5. The following packages were used: dplyr (v1.1.4), ggplot2 (v3.5.1) … Reproducibility auditing (reproducr) identified no risks. The full audit report and certification records are available in the supplementary materials.
"pharma"
A structured QC document with execution environment table, full package inventory, risk register, drift assessment, and sign-off fields for analyst and reviewer.
repro_report(report, risks, drift,
format = "html", style = "pharma",
output_file = "qc_report.html")CI/CD integration
reproducr is designed to run automatically on every push. A typical workflow:
- Audits your scripts for risk
- Checks for drift against the last certified run
- Updates the reproducibility badge in your README
See the reports and badges vignette for the complete GitHub Actions workflow.
Contributing
Breaking-changes database
The database powering risk_score() is maintained in a dedicated community repository — reproducr-db.
Contributing a new entry requires:
- A
pkg::fnkey - A version window (
from_version,to_version) - A risk level (
"high","medium", or"low") - A plain-English description of the breaking change
- A URL reference (package
NEWS.md, CRAN page, GitHub release)
Each entry is a small JSON file — see the reproducr-db README for the format and the contributing guide for the version window design principles.
Code of conduct
Please note that this project is released with a Contributor Code of Conduct. By participating you agree to abide by its terms.
Package development
To contribute to reproducr itself — new features, bug fixes, or additional risk checks — open an issue or pull request on the main repository.