A data quality case study from English local elections reveals how a categorical normalization bug completely reversed the headline findings. Analysis showed volatility nearly doubled from 12.0 to 22.5 between 2018-2022, yet treating similar party labels as separate parties incorrectly suggested widespread fragmentation. This data error demonstrates how raw strings can distort analytical categories if not properly normalized before computation.

The error came from treating ballot labels like "Labour Party" and "Labour and Co-operative Party" as distinct entities. After normalizing party families, the corrected data showed fragmentation increased in only 18 of 67 councils, not 66 as initially reported. Median fragmentation actually decreased slightly (-0.31), revealing the vote moved within an already-consolidating party system rather than splintering.

This case serves as a cautionary tale for data scientists working with categorical information. The principle extends beyond elections to product categories, job titles, or any domain where derived metrics depend on messy raw data. Always normalize categories before aggregation, or risk building an entire analytical framework on a foundational error that propagates through every downstream metric.