Reproducibility

This page gives reviewers the command path for checking the observatory snapshot, understanding live-data dependencies, and avoiding human-gated OBS-T annotation steps during unattended reproduction.

Local Layout

Full OBS-T checks expect this sibling layout:

GitHub/
  csl-observatory/
  csl-orig/
  CORRECTIONS/

Check it first:

python scripts/check_workspace.py

Read-Only Check

Use this path to verify an existing snapshot without intentionally refreshing generated data:

python scripts/refresh_observatory.py --check-only
python scripts/obs_t_regression.py
python scripts/workflow_health.py --check
python scripts/data_index.py --check
python scripts/repo_health_regression.py

Full Refresh

Use this path when refreshing generated reports and dashboard data:

python scripts/refresh_observatory.py --dry-run
python scripts/refresh_observatory.py

The default refresh reads live GitHub metadata and workflow metadata through gh api. If credentials are unavailable and placeholders are intentional:

python scripts/refresh_observatory.py --offline-metadata --offline-workflows

Offline metadata/workflow placeholders are useful for schema checks, but should not be cited as complete live coverage.

Dashboard Build

cd observatory\site
npm ci
npm run build

OBS-T Pipeline

The citable OBS-T release is guarded by:

python scripts/obs_t_regression.py

The full regeneration path is:

python scripts/build_correction_events.py
python scripts/reconstruct_git_events.py
python scripts/attribute_components.py
python scripts/attribute_crosswalks.py
python scripts/obs_t_release.py
python scripts/obs_t_typology.py
python scripts/obs_t_baselines.py
python scripts/obs_t_rigor.py
python scripts/obs_t_robustness.py
python scripts/obs_t_campaigns.py
python scripts/obs_t_translit_check.py
python scripts/obs_t_silver.py
python scripts/obs_t_issuelabel.py
python scripts/obs_t_regression.py

Human-Gated

Do not run these in unattended automation:

python scripts/obs_t_gold.py --make
python scripts/obs_t_gold.py --score
python scripts/obs_t_errorsample.py --make
python scripts/obs_t_errorsample.py --score

They create or score annotation sheets and require an intentional human review workflow.

Data And Citation

Use Data Downloads or observatory/site/src/data/data_index.csv to identify the exact public CSV/JSON file, generated date, source script, and caveat. Cite the repository commit or frozen tag together with the file and generated date. No Zenodo DOI minted yet — see Data Downloads.

The longer repository guide is docs/REVIEWER_REPRODUCIBILITY.md.