What 'let it emerge' actually produces

Collected examples from the last two years of systems I've worked with that were designed using a "let the structure emerge" approach. Meaning: no upfront schema, no naming conventions, no taxonomy. Build first, organize later or never.

Parts database. Three teams entering parts independently. 14 duplicate entries for a single bearing. Each entry uses a different description format. No one noticed for eight months because search returned results — just not consistent ones.

Network drive. ~2TB across 30,000+ files. Approximately 6,000 are duplicates at different revision levels. No way to determine which is current without opening each one and checking the content. Someone built a spreadsheet to track which files were the "real" ones. The spreadsheet is now also out of date.

ERP item codes. Started with a convention. Convention was not documented. New hires invented their own conventions. Now there are four coexisting systems for generating item codes. Two of them produce codes that are valid in the other's format, meaning you can't tell which convention a given code follows without looking up when it was created and by whom.

Project management tool. Tags were freeform. After 18 months: urgent, Urgent, URGENT, high-priority, high priority, priority-high, asap, and !!! all coexist. Filtering by priority is not possible.

In each case the fix was the same: stop, define a schema, migrate existing data, enforce going forward. The migration cost more than the initial schema would have.