SSH, Actually 100% signal, 0% noise.

Recovery Planning

concept
depth 3/3 − deeper + surface

Every authentication setup should be built with its own failure in mind: keys get lost, disks die, passphrases and PINs get forgotten. Recovery planning means deciding—before anything breaks—how you will get back in. Security you can't recover from isn't strength; it's fragility with good cryptography.

The critical safety rules:

  1. Always have a recovery path

    • Second hardware key with duplicate keys, OR
    • Disk-based backup key stored securely, OR
    • Out-of-band access to servers
  2. For resident keys: two hardware keys, always

    • Primary key for daily use
    • Backup key stored safely with duplicate resident keys
  3. Blobs aren't secrets, but their loss is permanent

    • Back them up
    • They can't be reverse-engineered
    • But losing them means generating new keys
  4. Test your recovery process before you need it

    • Can you regenerate stubs?
    • Can you access servers with backup key?
    • Do you know your PIN?

For disk-based keys: always use strong passphrases (20+ characters), store the passphrase in a password manager, and back up private keys to encrypted storage. For hardware keys: store the backup key safely (not with the primary), test the backup regularly, and document your recovery process.

Recovery planning is the practical face of a question that runs through the whole graph: where does authentication state live, and what happens when that place is destroyed?

  • A disk key's state is a file—its recovery story is backups (encrypted) plus the ability to revoke fast when the file falls into the wrong hands.
  • A non-resident hardware key splits state between device and blob—its recovery story is blob backups plus an accepted cost of regenerating keys if the device dies.
  • A resident key puts all state in the device—its recovery story must be a duplicate device, because nothing else can reconstruct it. This is why the resident-vs-non-resident choice is really a recovery-planning choice.

The supporting hygiene compounds here: one key per purpose (don't reuse keys across contexts—easier to revoke when needed), regularly audit authorized_keys on servers (remove old and unknown keys), and name keys clearly (makes revocation easier). Each of these turns a future emergency from an investigation into a checklist. And the untested plan is the one that fails: a backup key you've never authenticated with, a blob backup you've never restored, a PIN you think you remember—none of those are recovery paths yet. Test them while nothing is wrong.