Operating Playbooks
Governance · Operating overview · Inventory →
How-to for the Operating blueprint. These playbooks make the estate control system runnable: inventory, mapped obligations, and an evidence pack before scale. They are not PEP, SARAC, or five trust boundaries. Those live in Runtime.
Approve before scale. No inventory, no owner, no go-live. Runtime guardrails without this path is an agent with PEP and no examiner pack.
Three playbooks
| # | Playbook | What you build |
|---|---|---|
| ① | Inventory record | Named use case, owner, pattern, data class, go-live state |
| ② | Obligation-to-control map | Must-we → named control + evidence artifact |
| ③ | Go-live evidence pack | What examiners get before scale |
Recommended path
- Inventory: every in-production agent has a row before it calls a tool
- Obligation-to-control: each must-we maps to a control you can point at
- Evidence pack: assemble the pack; then Runtime for the request path
Worked example: bank refund agent
Same use_case_id on every artifact. If the id drifts, the examiner pack is fiction.
| Playbook | What exists for uc-refund-agent |
|---|---|
| Inventory | Row: Payments owner, Pattern 3 (guided hybrid), customer_financial, AU residency, status approved until the pack lands |
| Obligation map | Four must-we rows (operational resilience, privacy, consumer duty, internal fraud limit), each with a control_id you can open |
| Evidence pack | ev-refund-agent-2026-03: sample ALLOW / STEP_UP / DENY, eval suite green, named residual accept |
Then status may move to live. Runtime owns the request path: PEP on issue_refund, HITL above the fraud band, verdict log.
A staff HR policy FAQ is the same three playbooks with a thinner row: Pattern 0, internal_unclassified, no write-tool PEP. Do not copy the refund pack onto that use case.
What "not done" looks like
| Symptom | Missing playbook |
|---|---|
| Refund bot in Slack; nobody can name the owner | Inventory |
Legal memo "we comply with CPS 230" and no control_id | Obligation map |
| Screenshot PDF assembled the week before the auditor | Evidence pack |
Inventory live, tools called without PEP | Runtime (not these playbooks) |
Blueprint: Approve before scale. Canonical model: G.A.I.N Governance. Org planes: G.A.I.N AIOM.