From jle-skills
Assembles data and code package for JLE manuscript replication policy. Builds deposit and README; does not run analysis or write paper.
How this skill is triggered — by the user, by Claude, or both
Slash command
/jle-skills:jle-replication-packageThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- The paper is empirical (or has simulations/experiments) and you are heading toward acceptance, or want to build the package early
JLE's stated policy is to publish empirical/simulation/experimental papers only if the data are clearly and precisely documented and readily available to any researcher for replication, and authors of accepted papers must provide the data, programs, and other computation details sufficient to permit replication, prior to publication (检索于 2026-06;以官网为准; verify on journals.uchicago.edu/journals/jle/data-policy). This is a pre-publication requirement, not an acceptance-day formality. JLE does not run the AEA-style openICPSR pipeline; the obligation is documentation and availability — so build a package any researcher could re-run, and treat it as a gate, not a chore. Law-and-economics data raise distinctive access issues (court records, sealed cases, regulatory filings, proprietary market data) that you must plan for explicitly.
| Component | Requirement |
|---|---|
| Data files | All data used to produce the results, documented; or, for restricted data, a precise access path |
| Analysis + transformation code | Every script from raw legal/regulatory data → cleaned data → each table/figure |
| Master script | One run_all that regenerates every exhibit from the inputs |
| README | Data sources and provenance, computational requirements, run instructions, and an exhibit-to-code map |
| Data-availability statement | Provenance and access terms for each dataset (court system, agency, vendor), stated in the paper |
| Legal-data documentation | How statutes/cases were coded, the coding protocol, and inter-coder checks for hand-coded doctrine |
version + recorded ssc/net package versions; requirements.txt/conda env (Python); renv.lock (R).code/05_did.do, Figure 1 → code/06_event_study.R, etc.Adapt the vendored skeleton in
../../resources/code/(master script → clean → descriptive → DiD/IV/RD → mechanism → robustness → tables) as the package backbone.
run_all master script regenerates every table and figure from the inputsLaw-and-economics data come from a handful of recurring sources, each with its own access and documentation pattern. Name yours and document accordingly:
| Source | Access reality | What to document |
|---|---|---|
| Court records (PACER, state dockets) | often public but fee-gated or rate-limited; some sealed | the query/scrape procedure, date pulled, sealed-case handling |
| Administrative / regulatory filings (SEC EDGAR, agency dockets) | usually public | the form types, vintage, and any parsing code |
| Enforcement / litigation databases (vendor) | proprietary, license-restricted | the vendor, license terms, and a synthetic extract |
| Statute / case coding (hand-built) | you create it | the coding manual, sources, and inter-coder reliability |
| Linked administrative microdata (sealed) | DUA-restricted, non-redistributable | the application path, DUA terms, wait time, synthetic schema |
The reproducibility obligation is the same in every row: someone who legitimately obtains the source must be able to re-run your code and recover every exhibit.
A paper on judge assignment uses individual case records the court provides only under a data-use agreement. The author cannot redistribute them, so the package deposits: (i) all cleaning and analysis code; (ii) a documented access path (the court's data-request form, the DUA terms, the ~10-week wait); (iii) a synthetic case file with the same schema so run_all executes end-to-end and a verifier can confirm the logic; and (iv) the judge-leniency leave-out construction script. The hand-coded ruling-type variable ships with its coding manual and a 200-case inter-coder reliability table. Any researcher who obtains the records can reproduce every exhibit — the JLE standard.
【Master script】run_all regenerates all exhibits from inputs? [Y/N]
【Data】documented + available, or restricted path + synthetic extract? [state]
【README】exhibit-to-code map + computational requirements complete? [Y/N]
【DAS】provenance + access terms for every dataset? [Y/N]
【Legal coding】protocol + sources + inter-coder reliability deposited? [Y/N/NA]
【Reproducibility】versions pinned + seeds + no absolute paths + clean fresh run? [Y/N]
【Restricted/sealed】flagged early? [Y/N/NA]
【Next step】jle-referee-strategy (or jle-submission)
npx claudepluginhub brycewang-stanford/awesome-journal-skills --plugin jle-skillsBuilds a reproducible replication package for JLEO manuscripts from institutional data sources to every exhibit, including documentation of data construction and environment pinning.
Guides authors of accepted JOLE papers through assembling a data/code deposit for the JOLE Dataverse Repository, including proprietary-data handling and the AER data-availability policy.
Assembles a JDE-compliant replication package (data, code, README, pinned environment) for submission or acceptance. Does not run the analysis.