Research runs on a local-first stack: scheduled collectors feed a Postgres and Redis data layer, dashboards and monitors watch it, and multi-tier backups are exercised by scheduled restore drills. Everything below is read from the machine itself by an inventory script, not written by hand.