Philip Sagalovich

• Getting off a managed platform. Supabase, Firebase, Heroku. Same schema, same policies, same front-end code, running where you decide.
• Postgres that got slow. Usually the config nobody touched since install, and two or three queries doing full table scans because the planner was priced for spinning disks.
• Backups you have actually restored. Encrypted, offsite, and tested — rather than a cron job nobody has ever run backwards.

The Supabase case is written up in detail and has its own page with scope and price.