Review R Code

Run a code review protocol on an R script or QMD chapter.

Input: $ARGUMENTS — path to .R or .qmd file.

Review Dimensions

1. Reproducibility

• set.seed() present if any randomness used
• All paths are absolute or use defined root variable
• library() calls at top (not scattered)
• No hardcoded values that should be variables
• Data loading is deterministic (no web scraping without caching)

2. Correctness

• Regression specification matches paper ({{unit_fe}}+{{time_fe}} FE, clustered SEs)
• Missing value handling is explicit (na.rm=TRUE where needed)
• Joins preserve expected row counts (check for accidental duplication)
• Factor/character conversions are intentional
• Variable names match Stata equivalents for cross-verification

3. Conventions (from .claude/rules/stata-r-conventions.md)

• library() not require()
• |> preferred over %>%
• fixest::feols() for TWFE
• modelsummary for tables
• haven::read_dta() for Stata files
• snake_case naming
• Comments explain non-obvious logic

4. Quarto-Specific (if .qmd)

• Every code chunk has a unique label:
• echo: and eval: set appropriately
• Table/figure chunks have captions (tbl-cap:, fig-cap:)
• Cross-references use @sec-, @tbl-, @fig- syntax
• No orphaned code chunks (every chunk has surrounding narrative)

Output

Report by severity (do NOT edit files, report only):

• Error: Will produce wrong results or fail to run
• Warning: May produce unexpected behavior
• Style: Convention violation, should fix for consistency
• Note: Suggestion for improvement

Format:

[SEVERITY] Line [N]: [Description]
  Found:    [what's there]
  Expected: [what should be there]