Principle: Single source of truth

When the same information exists in multiple places, they inevitably go out of sync. Bugs arise from using the stale copy.

Every piece of knowledge should have one authoritative source:

1. Data: One database table owns each entity

Bad: user email in users table AND in orders table
Good: orders references users.id, join for email

1. Configuration: One config file, derived elsewhere

Bad: version in package.json AND in docker-compose AND in CI
Good: Read version from package.json everywhere

1. Types: Generate from schema, don't duplicate

Bad: API types defined in frontend AND backend
Good: OpenAPI spec generates types for both

1. State: One component owns each piece of state

Bad: Same state in parent and child via two-way binding
Good: Lift state up, pass down as props

1. Documentation: Code as source, docs generated

Bad: README manually duplicates code behavior
Good: Tests document behavior, API docs auto-generated

When caching or denormalizing (read replicas, CDN, search index):

• Acknowledge these are copies, not sources
• Have a clear invalidation strategy
• Document which is the source of truth

Duplicated knowledge creates maintenance burden and inconsistency. Changes must be made in N places instead of 1, and forgetting any creates bugs.