Set up a proper migration system for my database. Check what I'm using now: an ORM with auto-sync, manual SQL edits against the live database, or nothing (the schema just exists because the code expects it). Implement: versioned migration files that run in order, each with an up and a down path; a tracking table recording which migrations have run, so running them twice is a no-op and new environments catch up automatically; migrations wired into my deploy process so schema changes apply before the new code starts; and my current schema captured as the baseline migration so a fresh database (new developer, staging, disaster recovery) can be built from nothing. Disable any ORM auto-sync or 'alter tables to match models' behavior in production; it's how columns get dropped silently. Write migrations to be deploy-safe: adding columns with defaults, creating indexes concurrently where the database supports it, and avoiding table locks on large tables. Never have a migration destroy data without me explicitly approving it in the file. Deliver: the migration setup, the baseline, the deploy integration, and a verification: build a fresh database from scratch using only the migrations, and roll back the last one to confirm the down path works.