Decision records

The calls that are expensive to get wrong, with an actual answer.

Every article about these ends in “it depends”, which is true and useless. Each of these commits to a call, says when the other option is right, and then tells you the part nobody writes down: what the choice costs you eighteen months later.

6 records

• Supabase or your own Postgres?Data
  When to stay on Supabase, when to run your own Postgres, and the row-level security trap that turns the choice into a data leak.
• Next.js route handlers or a separate backend?Architecture
  When API routes are enough, when to split out a service, and why where you put the business logic matters more than which you choose.
• Buy authentication or build it?Security
  Why buying auth is almost always right, the two narrow exceptions, and the one hour of work that decides whether you are locked in.
• A queue or a cron job?Background work
  Cron for work the clock starts, a queue for work a person starts, and the two failure modes that catch everyone who guesses wrong.
• Monorepo or separate repos?Repository
  Why one repo is usually right, why that is not the same as adopting a monorepo toolchain, and the version-skew bug that decides it.
• When do tests start paying for themselves?Quality
  Not on day one, and earlier than most founders think. The real trigger, and why agent-written code changed the answer.