Clean a messy spreadsheet reproducibly

Here's a messy export: {{paste 20-30 sample rows with headers}}.

1. Diagnose every problem you can see: inconsistent date formats, numbers stored as text, mixed units, trailing whitespace, merged categories, duplicate entities under different spellings, currency symbols, thousands separators, and rows that are actually subtotals.
2. Propose the cleaned schema with types and canonical value sets.
3. Write the transformation as code in {{Python/SQL}} — deterministic, re-runnable, with each step commented as a rule.
4. For fuzzy work like name deduplication, show the matching rule and print the pairs it merges so I can review rather than trust.
5. Add validation assertions that fail loudly: row count expectations, no nulls in key columns, totals matching source, no unexpected category values.
6. List the decisions where you had to guess my intent, and what I should confirm.

Never silently drop rows. Everything excluded goes to a rejects table with a reason.