Staging runs on production data that never left the disk

Staging needs realistic data. The usual answer is a nightly dump to a bucket, an anonymization script that's three schema versions behind, and a download onto whichever laptop is running the migration test.

Instead: snapshot the production container, clone it into a staging container on the same bare metal server. The clone shares disk blocks copy-on-write, so a 40 GB database is ready in seconds and costs almost nothing idle. The SQLite file under /hoody/databases/ is there exactly as production had it, one minute ago.

No bytes crossed the internet. No object store holds a dump. No laptop has a copy to lose. When the test finishes, delete the clone and the data is gone with it.

The copy is on the same disk as the original; the only thing that moved is a pointer.