Performance comes after model quality. Start with correct entities, keys, and constraints, then add indexes to the hot paths and denormalize only when measured workload pressure justifies it.

Index primary keys and foreign keys used in joins.

Keep rows tight so scans and joins stay cheaper.

Use transactions carefully on hot write paths.

Measure join and index behavior as the dataset grows.

Duplicating the same business fact across multiple tables.

Skipping foreign keys and trusting application code alone for integrity.

Treating NoSQL as a universal replacement instead of a workload decision.

Adding schema complexity before the workload justifies it.