Planned
AI provider outage response
Identify the blast radius, invoke fallback behavior, communicate clearly, and restore service without hiding degraded quality.
Operator-ready procedures
Station 06
Runbooks
When the system is wobbling, start here.
Runbooks turn a tense moment into a sequence of responsible actions. They are written for the operators who need to stabilize the service, protect the client, and preserve the evidence.
On the prep list
What we’re working on.
These are the first topics on the board. We’ll link each item here when it is genuinely ready to use.
Planned
Unsafe output or data exposure
Contain the workflow, preserve evidence, assess impact, notify the right owners, and define safe re-entry criteria.
Planned
Model or prompt change rollback
Detect regression, compare known-good behavior, restore the prior configuration, and document the decision.
How we’ll serve it
Calm, explicit instructions for consequential moments.
Runbooks will define triggers, owners, immediate actions, escalation paths, validation, communications, and the conditions required to close the incident.
- 01Clear triggers
- 02Named decision points
- 03Closeout criteria