Reading a Snowflake schema the way a new analyst would

A new analyst does not read your schema top to bottom. They find one table they trust, follow it outwards and build a map as they go. That turns out to be a good algorithm.

Start where the volume is

The tables that get queried most are the ones people already trust. They are the right place to anchor a map, and the query history tells you which they are without asking anyone.

Names carry more than they should

A column called org_id in one schema and company_id in another is usually the same thing, and a column called id is usually not. Names are evidence, not proof, so every inferred link keeps the reason it was made.

Ask about the leftovers

Every warehouse has a table that looks important and is dead, and one that looks like a scratch table and runs payroll. That knowledge lives in people’s heads and in tickets, which is why reading docs and call notes alongside the schema is not a nice extra.