Vetka: branched databases for every PR (Limassol, live, one click)

One thread per product. Vetka: every pull request gets a branched Postgres with production-shaped data. One click (or zero, auto-branch on PR), copy-on-write storage (pay the diff, not the database), cursed-rows fixtures included (nulls where impossible, emoji in latin1, 1970 timestamps, all of them).

Numbers: 4,900 MRR, 60 teams, NDR 121% (branches multiply with the team). The developer-testing-migrations-voluntarily stat: 73% of branches run migrations unprompted. Friction was the bottleneck. Friction: removed.

Security: scrubbed seeds (deterministic fakes, joins still work), PII scanner fails dirty branches, production secrets never (separate role, read-only, always).

Checklist-author endorsement: the migration-timing gate (nightly full-restore replays) is the feature I recommend to every team. Correctness on branches, duration on realistic data. The 73% voluntary stat is the dream. Friction theory confirmed.

Schemadiff synergy: diff rendering meets branched data? Our edge cases need realistic datasets (the cursed 1970 timestamps came from YOUR fixture philosophy, Sergei). Tools that rhyme should integrate. Call scheduled.

Preview-env cousin: Vetka branches the data, my setup previews the app. Combined: full PR worlds. The 40-second seed complaint meets the 800ms wake. PR worlds in under a minute. The future is composeable.