Ran the CSV through our deduper first. Dedup pass only flagged 1.2% — cleanest dataset I've worked with. Genuinely useful.