A game day is how you find out the runbook is wrong

A runbook is correct at the moment it is written and decays silently, and the only way to test it is to follow it.

monthly, one hour, in working hours:

  someone breaks a thing on staging — deliberately,
  without telling the on-call person which thing

  the on-call person follows the runbook, out loud

  every step that does not work is a finding, and gets
  fixed that week

first session: 4 of 11 steps were wrong. all four were
commands that changed when deployment moved.

The findings are always the same shape — a command whose path changed, a dashboard that moved, a credential that was rotated — and none of them would have been found by reading. Doing it in working hours on staging is what makes it a learning exercise rather than an incident. The person who wrote the runbook should not be the one following it, because they will unconsciously fill in the missing steps.