Write a data dictionary

What it asks you for

Your tablesrequired

Tables and columns, however you have them.

Known traps

The things that burn people.

What it tells the AI to do

Document this schema so queries stop guessing.

{{ schema }}
{% if traps %}
Known traps: {{ traps }}
{% endif %}
For each table: what one row means, in a sentence. That sentence is the single
most useful line in a data dictionary and the one most often missing, because
it is where "one row per order" and "one row per order line" get confused.

Then per table: the columns that matter, their units, their nullability, and
what a null actually means there. A nullable `country` that means "unknown"
and one that means "domestic" produce different numbers from the same query.

Record the conventions once: timezone, whether money is in cents, how soft
deletes work, which ids are test accounts.

End with the traps: joins that fan out, columns that look like a metric but are
not, and any table where the obvious query gives the wrong answer.

• data
• schema
• documentation
• sql

More data skills

• Extract to a tablePull structured rows out of documents, pages, or notes.
• Analyse a spreadsheetLoad a CSV or spreadsheet and answer questions about it.
• Chart this dataTurn numbers into one chart that makes the point.
• Clean a datasetDeduplicate, normalise, and repair a messy table.
• Read out an A/B testCall a result, including when the answer is that it is too early to call.