archive

/mstack:archive

Stage: reflect (at acceptance) Voice: replicator

When to invoke

The paper is accepted. Before the journal sets the production deadline you forgot about, lock the replication package. The point is not to satisfy the journal's data policy — it's to ensure that two years from now you (or anyone else) can rebuild every table and figure from raw data.

Procedure

1. Load context.

   • .mstack/config.yaml for paper title, target journal (now the actual journal).
   • paper/, data/, code/, output/.
   • prereg/ if applicable.

2. Inventory. Walk the project and list:

   • Raw data files in data/raw/ with sources, vintages, licenses.
   • Cleaned data file(s) in data/clean/.
   • Code files in code/, ordered by run order.
   • Output files in output/ (tables, figures, model objects).
   • Manuscript files (final accepted version).
   • Preregistration documents.
   • .mstack/llm-usage.jsonl plus every prompt_user_ref / code_ref file it names (GUIDE-LLM artifacts), if the ledger exists. Save the inventory to replication-manifest.txt.

3. Check licenses. For each raw data file:

   • Public-domain or open: include in the package.
   • Restricted (e.g., commercial, IRB-restricted): replace with a stub + script that re-acquires from source. Document the restriction in the README. Tag each file accordingly in replication-manifest.txt.

4. Pin dependencies.

   • R: capture sessionInfo() and (preferred) generate a renv.lock via renv::snapshot(). Add renv/activate.R so the package self-restores.
   • System dependencies: list in the README (e.g., LaTeX distribution, JAGS, Stan, GDAL).
   • Quarto/LaTeX versions if relevant.

5. Extend README.md into a replication README with:

   • One-paragraph paper summary + final citation.
   • Layout (the standard MStack tree).
   • Reproduction steps — run every numbered script, not a hardcoded four; /mstack:codebook, /mstack:power-analysis, and /mstack:robustness add 00/05/06-series scripts whose outputs also ship:

     renv::restore()
     scripts <- sort(list.files("code", pattern = "^[0-9].*\\.R$", full.names = TRUE))
     scripts <- scripts[!grepl("00-fetch", scripts)]  # raw data ships in the package; fetch scripts hit the network
     for (f in scripts) source(f)
     

   • Mapping: which script produces which table / figure (a table indexed by output/ filename).
   • Data source documentation (one entry per raw file).
   • Known divergences from the published paper (e.g., a figure that was hand-edited in Illustrator — flag the post-hoc edit).
   • License (CC-BY for code unless user specifies otherwise; data licenses inherit from sources).
   • Contact + DOI placeholder for the published paper.