The policy should be short enough to execute and strong enough to change behavior.
Release normally. Keep routine reliability work on the roadmap, but do not slow product flow unnecessarily.
Require extra review for risky changes, investigate burn rate, and prioritize the most likely stability wins.
Freeze non-essential risky releases, focus engineering effort on stabilization, and only resume normal pace when the service is back inside policy.
Why this works
It removes improvisation in the middle of an outage or release debate. The team knows in advance what the data means operationally.