Seed data

Seed data is the set of records a script inserts so a fresh database is actually usable. It comes in two flavors that people constantly conflate. Reference seeds are rows the product cannot run without: roles, plan tiers, currencies, country codes. Production needs these. Development seeds are believable junk: fifty fake users with Faker names, a few orders, one account already mid-onboarding, so you can see the UI in a realistic state instead of staring at empty states all day.

Write the seed script to be idempotent: upsert on a stable natural key rather than blindly inserting, so running it twice does not duplicate everything. Fix the random generator's seed value so the same command produces the same data on every machine, which turns 'works on my laptop' bugs into reproducible ones. Most stacks ship a hook for this (prisma db seed, drizzle-kit, rails db:seed, a plain SQL file run after migrations).

The classic trap: dev seeds quietly becoming production data. Someone hardcodes id 1 as the admin, prod gets seeded once, and now real code depends on a fake row. Keep the two seed sets in separate files with separate commands. Second trap: seeds that bypass your app's validation and create states your real code can never produce, so you debug phantom bugs.