Ask HN: Best Practices for Data Dictionaries?

We're tackling the goal of "improving our shared understanding of data sources and purposes", one element of which is a data dictionary.

This could mean a lot of different things ~ comments on all source database columns? Annotations of data warehouse tables? A centralized document with high-level definitions of concepts like "user" or "pilot"?

Questions we are facing include:

- Who is responsible for what activities?

- How do we keep it up to date?

- Built into source systems, or separate?

- What format?

Any hard-won advice or speculative suggestions?