Can a reviewer rerun your R result?
Sharing code is not the same as sharing something that runs. Missing package versions, an unstated R version, or an undeclared system library are the usual reasons a rerun fails. This check finds them before a reviewer does, and tells you what to fix first.
What we check
R version
Is it recorded, and can that version still be installed?
Packages and lockfile
Is every package declared, pinned, and in sync with the code? Supported first: renv.
System dependencies
Are the libraries behind packages like sf, rJava, or xml2 documented?
Quarto / R Markdown render
Do the paths, seeds, inputs, and render command work outside your machine?
What we rerun
Together we agree on one output: a table cell, a coefficient, a figure. We rebuild your environment from what the project declares, in a disposable container, run it, and compare the result to the value in your manuscript.
What you get
- Pass/fail checklist with the evidence for each item.
- Captured session and dependency manifest from the rerun.
- Output comparison: expected value, observed value, and the tolerance used.
- Prioritized fixes in plain language, ordered by what unblocks a rerun first.
Delivered as a PDF and Markdown packet. Every failed rerun is reviewed by Nedim Yel, PhD, before anything is sent, with the evidence and limitations documented.
Limits, stated plainly
- This is a pilot. Spots are limited, and scope is set per project while the check is tested on real repositories. We confirm terms with you before any work starts.
- One output per check, one environment (Linux container). It does not assess whether the analysis is statistically sound; that is the manuscript review.
- renv is supported first. Notes on rix, uvr, or single-file approaches are informational only, not tested by us.
- Please do not send sensitive or participant-level data. A sample project with synthetic or public data is enough. Files are deleted within 30 days; see the privacy policy.
Need to check the numbers across your paper?
The Manuscript Results Check traces up to 80 agreed manuscript claims to their R or Python outputs, with human review of group labels, rounding and unsupported claims.
Want to be in the pilot?
Choose Reproducibility Check in the quote form and tell us about your project.