Maintainer setup

replicateEverything separates readers (browse, Run) from maintainers (onboard a study or server). Live Run and Shiny probe dependencies only — they never install R, Python, or Stata packages on the host. Maintainers install once during setup.

library(replicateEverything)

Check system compatibility (no installs)

Before running tables or figures, confirm the machine matches what replication.yml declares:

check_study_compatibility("10.1017/S0003055426101749")

The returned list includes ready (logical), install_needed, and per-engine blocks under dependencies (r, python, stata).

In Shiny, use Check system compatibility above the tables list. When something is missing, a dialog shows the same maintainer commands documented here.

Install dependencies for one study

Use one function for folder-backed and package-backed registry studies (R CRAN, Python pip, Stata install_stata_deps.do):

install_dependencies("10.1017/S0003055426101749")
install_dependencies("10.1371/journal.pone.0278337")  # package-backed too
install_dependencies("path/to/study-repo")

This does not rebuild display artifacts — only dependencies. After setup, build artifacts with the kind-specific builder (paths differ by design):

build_study_outputs("path/to/folder-study", install_deps = TRUE)
build_study_outputs("rep1371journalpone0278337", install_deps = TRUE)

Folder-backed studies write under outputs/; package-backed studies use inst/report/outputs/ (or legacy inst/report/artifacts/).

Install dependencies for every registry study

On a shared audit or Shiny server, run once:

install_dependencies("everywhere")

Covers folder- and package-backed entries. Failures are collected per DOI; other studies continue.

Point R at Python and Stata permanently

Session options() are not the right place for production servers. Set environment variables in ~/.Renviron and restart R:

PYTHON=C:/Users/you/AppData/Local/Python/pythoncore-3.14-64/python.exe
STATA=C:/Program Files/Stata17/StataMP-64.exe

On Linux or macOS, use the corresponding paths (for example /usr/bin/python3 or /Applications/Stata/StataMP.app/Contents/MacOS/stata-mp).

In R you can open the file with usethis::edit_r_environ() if you use usethis.

Priority order:

Engine	Permanent config	Fallback
Python	PYTHON or RETICULATE_PYTHON in .Renviron	Windows py -0p installs, then PATH
Stata	STATA or REPLICATE_STATA_EXECUTABLE in .Renviron	options(replicateEverything.stata_executable), then common install paths

What happens on Run

When a reader calls run_replication() or Shiny Run, the package:

1. Checks declared dependencies (assert_study_ready_for_replication()).
2. Stops with [maintainer_dependency_hint()] if anything is missing.
3. Runs the replication without installing packages.

Maintainers see the same hint text in the console, in Shiny modals, and via maintainer_dependency_hint():

maintainer_dependency_hint("10.1017/S0003055426101749")

See also Meet the functions, the folder replication checklist, and inst/ai/skills/include_study_in_registry.md for contributor vs maintainer registry workflows.

Register studies in the central registry (maintainer)

There is no study-local registry handoff — not registry/ in a folder-backed repo, not inst/registry/ in a package. Contributors validate with [check_and_bake_study()]; a maintainer with a local checkout of the registry repository writes the stub directly from the study’s own replication.yml with [sync_study_to_registry()] — nothing is copied from the study repo.

Point R at your monorepo checkout:

options(replicateEverything.registry_root = "../registry")

Sync one study — build registry/studies/<folder>.yml from the study’s root yaml and rebuild index.csv:

sync_study_to_registry(
  "../rep-10.1177-00491241211036161",
  registry_root = "../registry"
)

Optional per-study audit after sync (upserts that DOI into audit_jobs.csv; does not wipe the portfolio health bar):

sync_study_to_registry(
  "../rep-10.1177-00491241211036161",
  registry_root = "../registry",
  audit = TRUE,
  patience = 20
)

Refresh the whole registry after a batch of syncs — recompile index.csv from all stubs and rerun [audit_everything()]:

refresh_registry("../registry", audit = TRUE, patience = 20)

Audit disk layout (as of 0.7.40): registry/audit_jobs.csv is the source of truth (upsert by doi × object × engine). Derived audit_summary.json / audit_latest.rds are rebuilt from the full CSV after each write. To fill gaps without live runs, use [seed_registry_audit_jobs()]; to rebuild JSON/RDS only, [refresh_registry_audit_summary()]. Commit audit_jobs.csv together with the derived JSON and RDS (and the Quarto HTML if you publish the registry report). See vignette("audit") and registry/guides/registry-management.md.

One call, check + sync: [register_study()] runs check_and_bake_study() then sync_study_to_registry() for both folder- and package-backed studies:

register_study(
  "../rep-10.1177-00491241211036161",
  registry_root = "../registry",
  audit = TRUE
)

Check precomputed outputs

After building display files, confirm every declared table and figure exists on disk (Shiny Display mode). This does not run live replications.

validate_outputs(location = "../rep-10.1177-00491241211036161")
validate_outputs("10.1177/00491241211036161", what = "everything")

Registry-wide check from a monorepo checkout:

options(replicateEverything.registry_root = "../registry")
validate_outputs(doi = "everywhere", what = "everything")

Or from the registry repo: Rscript scripts/validate_outputs.R.

Summary

Task	Function
Probe this machine	check_study_compatibility(doi)
Install one study (all kinds, all languages)	install_dependencies(doi)
Install entire registry	install_dependencies("everywhere")
Build study outputs	build_study_outputs(location, install_deps = TRUE)
Validate study	check_replication(location)
Check precomputed outputs	validate_outputs(location) or validate_outputs(doi, what = "everything")
Registry-wide output check	validate_outputs(doi = "everywhere", what = "everything")
Hint text for errors / UI	maintainer_dependency_hint(doi)
Contributor: validate (+ optional bake)	check_and_bake_study(path)
Maintainer: sync stub into registry	sync_study_to_registry(path, registry_root = ...)
Maintainer: validate then sync in one call	register_study(path, registry_root = ...)
Maintainer: rebuild index + audit all	refresh_registry(registry_root, audit = TRUE)
Rebuild index only	build_registry_index(registry_root)
Seed audit CSV (no live runs)	seed_registry_audit_jobs(registry_root)
Rebuild audit JSON/RDS from CSV	refresh_registry_audit_summary(registry_root)

Package website (pkgdown)

The site is served from docs/ on main (GitHub Pages → /docs). CI does not build or deploy it.

# from replicateEverything/
Rscript scripts/build_pkgdown.R
# then: git add docs/ && git commit && git push

Commit the entire docs/ tree (deps/, articles/, reference/, …). Pushing only index.html breaks Bootstrap styling on GitHub.

Windows + Dropbox: if the repo lives under Dropbox, pkgdown may fail while writing docs/articles/*.html (e.g. audit.html) with Invalid argument [1515] / Error closing file. Pause Dropbox, exclude docs/ from selective sync, or clone and build outside Dropbox. scripts/build_pkgdown.R deletes cached article HTML first when it detects a Dropbox path.