- Prefer curated character names over cached or generated translations
- Preserve name spellings inside compound labels and parsed speakers
- Reload vocab before runs and refine default event-code handling
- Add regression coverage for glossary precedence and matching
