Danger zone

Sharp edges when using open_fdd as a library.

---

1. Trust rule YAML like code

Only load rules from trusted paths. Expression rules run pd.eval / restricted eval over your DataFrame columns—do not merge untrusted YAML into production without review.

---

2. Column map mistakes look like “silent” skips

With skip_missing_columns=True (default), a bad column_map can cause a rule to be skipped with a warning instead of raising. Use input_validation='strict' and skip_missing_columns=False in CI when you want failures loud.

---

3. Units and scaling

Expressions use raw column values. 0–1 vs 0–100 command scaling is your responsibility unless you call normalize_cmd(...) in the expression (see Expression rule cookbook).

---

4. Operational data deletion

Deleting rows in your warehouse or database is outside this package. If you deploy a full platform, data-retention and CRUD delete semantics are documented in open-fdd-afdd-stack.