This blog’s whole philosophy is ship by feel, skip the ceremony, fix it when it breaks. It says so in the tagline. So consider this the fine print, the one clause of the vibe contract written in boring lawyer font: backups are exempt. You cannot vibe a backup.

Why Vibes Fail Here Specifically

Vibe-driven infrastructure works because most failures are recoverable. Container crashed? Restart it. Config wrong? Fix it. Deploy broke the site? Roll back. The feedback loop is fast and the blast radius is small — that’s what makes moving fast rational at home.

Data loss is the one failure with no feedback loop. There is no “fix it when it breaks,” because broken means gone. The photos, the documents, the databases with years of accumulated history — deleting the wrong volume takes four seconds, and no amount of skill afterward un-deletes it. Vibes are a strategy for recoverable systems. Backups are what makes the rest of the system recoverable.

The Three Lies Self-Hosters Tell Themselves

“I have RAID.” RAID is availability, not backup. Every mistake you make gets mirrored, instantly, with perfect fidelity, to all disks. Redundant copies of a deletion are still a deletion.

“I have snapshots.” Better — until the box holding the snapshots is the box that dies, gets encrypted by ransomware, or takes a power surge. Snapshots on the same hardware are a seatbelt in a car that’s on fire.

“I set up backups last year.” Did you restore one recently? An untested backup is a hypothesis. Backup jobs rot quietly: the target fills up, the credential expires, the new service you added in spring was never included. The job that reports success while backing up nothing is a genre classic.

The Minimum Respectable Setup

  • 3-2-1: three copies, two different media, one offsite. Offsite can just be encrypted cloud storage — the point is surviving a fire, a flood, or a burglar.
  • Automated. Anything requiring a human to remember will eventually not happen.
  • Monitored. Alert when a backup doesn’t run. Silence must mean success, and that has to be true.
  • Restore-tested. Quarterly, pull a real file and a real database from backup. Ten minutes. It converts hope into knowledge.

The Payoff

Here’s the part nobody advertises: good backups make the vibes better. Once restores are proven, you can experiment violently — upgrade the risky thing, refactor the storage, try the sketchy migration — because the worst case stops being catastrophe and becomes inconvenience. The most reckless-looking operators you know are the ones with the most boring backup regimes. That is not a coincidence. That is the trick.