History Migrations Era

Scope

A single commit on 2026-04-27. Late but significant — added the schema_migrations tracking table that should have existed from the start.

Co-author: Claude Opus 4.7 (1M context).

What landed:

migrations/000_schema_migrations.sql — bootstraps a tracking table:
- version PRIMARY KEY (TEXT).
- name nullable TEXT for legacy compatibility (some migrations insert without supplying a name).
- applied_at TIMESTAMPTZ DEFAULT NOW().
Renamed later-committed 007_add_archive_tables.sql → 009_add_archive_tables.sql to resolve a numbering collision with 007_pgvector_company_embeddings.sql (kept as 007).
New runner: scripts/migrate.ts. Idempotent. Uses Bun.sql for bookkeeping and psql -f for migration bodies. Supports --mark-applied <version> to backfill rows for migrations applied out-of-band before tracking existed.
Fixed migration 007 (the surviving 007, pgvector): removed an illegal BEGIN/COMMIT wrapper around CREATE INDEX CONCURRENTLY statements. Dropped a stray idx_bv_sni_primary that referenced a column that does not exist on bolagsverket_companies.
Wired bun run migrate in package.json.

Verification cited in body: “all 10 migrations apply cleanly against the live DB and a re-run is a no-op (applied=0 skipped=10).”

Before this commit the project had 10 migrations and no record of which ones had been applied. Anyone setting up a fresh DB had to read each *.sql and check by hand.
The numbering collision is a permanent record: if you git log for 007_add_archive_tables.sql you will find it under both filenames. This commit is where the rename happened.
The --mark-applied mode is the migration path for existing prod databases. Run it once per applied version to backfill the tracking table.