Write the data dictionary nobody wants to write

Here's a table: {{paste the schema, plus 10-20 sample rows}}. Team context: {{what the table is used for, who queries it}}.

For every column produce: the plain-English meaning, the type, the unit or currency, whether nulls are allowed and what a null actually means here, the realistic value range from the sample, and whether it's safe to aggregate.

Then add the parts people always leave out: which columns are populated by which system and when, which ones are updated in place versus append-only, the timezone of every date field, and the definition of any column whose name is ambiguous ({{e.g. 'status', 'active', 'value'}}).

Flag: columns where the sample data contradicts the column name, columns that look like they hold two different meanings, and any field that looks like personal data.

Finish with the three questions I need to ask the owning team before this dictionary can be trusted.