What's the worst Supabase migration you've ever done?

Building Twinstack came out of one of mine — three days of scripts, half a dozen half-broken pg_dumps, and one very confused staging environment. Since launching, the most-repeated story I hear is auth: someone runs a "clone" that copies the users table but not the password hashes, and their entire user base is silently locked out until Monday morning.

So — what's yours?

• Staging refresh that turned into a full-day quest?

• The one storage bucket you forgot until a customer complained?

• Regional migration where the RLS policies just… didn't come along?

• Or a happy path story I can learn from?

Curious to hear where the current tooling breaks down for you. I'm collecting these; they're what shape the roadmap.