Join Safety

When to use

Any time you change a join, filter, crosswalk, aggregation, or key definition.

Checklist

• State intended keys (left + right).
• Assert uniqueness on the side(s) expected to be unique.
• If many-to-many is possible: aggregate first OR build an explicit crosswalk.
• Log before/after row counts and the share of unmatched keys.
• Add a regression test or a small invariant check (e.g., totals within tolerance).
• Treat required schema columns as mandatory: if missing, stop(...) with a clear error.

Banned anti-pattern for required columns

• Do not use intersect(required_cols, names(dt)) + if (length(...)) { ... } else { fallback }.
• Do not silently substitute broad fallback columns (e.g., total_violations) when specific required columns are missing.

R helpers (preferred pattern)