List entities, list fields, pick types, add validation, add defaults, decide ref vs embed, add indexes, add timestamps, mark sensitive fields, write the schema file, add hooks, test, use in routes, and monitor in production.

After listing fields and types, before adding indexes. For each relationship, decide: ref for large N (messages, connections), embed for small tightly coupled data (an address in a user). This decision is hard to change later.

After entities, fields, types, validation, defaults, and ref vs embed. Index the queries you actually run. Use explain() to verify. Add more indexes as the data grows and query patterns emerge in production.