- How often
- Once per identity. Change the password if you ever suspect it leaked.
- How long
- Ten minutes, of which one matters far more than the other nine.
- What it prevents
- A credential pair leaked from an unrelated service being tried against you by something automated.
- If you skip it
- The most impersonal attack there is reaches you, and nobody had to target you for it to work.
- How you notice you stopped
- Not applicable, and the version that goes wrong is reusing a password later out of convenience.
Uniqueness rather than strength
Strength protects against guessing, which is not the attack. The attack is a credential pair leaked from somewhere unrelated being tried everywhere by something automated. A long complicated password reused somewhere that leaked it is exactly as useless as a short one, and the confidence it produces makes it worse.
What a second factor covers
| Attack | Stopped? |
|---|---|
| A leaked password tried later | Yes, completely. Exactly what it was designed against. |
| A live copy of the login page | No. It collects the password and the code together and replays both while the code is valid. |
| Something running on your own machine | No. Nothing on the account side helps. |
The minute people regret
- Save the recovery material at setup, before closing the page. Recovery without it is limited and frequently impossible.
- Keep it separate from the password. Both in one place is one factor wearing a costume.
- Do not keep the only copy on one device. Devices get lost and replaced, usually at the worst moment.
- Not in a screenshot. Photo libraries sync, back up, and get indexed by software that reads text in images.
- Test once that you can actually produce it from a second place.
Also worth deciding now: the name. Not one used anywhere else and not a variation on one you use, because variations are the easiest thing in the world to match. Unlike a password, a handle cannot be changed later without abandoning whatever record is attached to it.