# Network Change Evidence & Rollback Handoff Desk

Use only fictional or expressly approved, non-sensitive training inputs. Do not paste configurations, device names, IP addresses, customer data, credentials, keys, monitoring screenshots, topology files, or incident details into an AI tool unless a named owner has approved that exact system and data flow.

## 1. Authority receipt

- Change owner:
- Architecture reviewer:
- Operations/change authority:
- Security reviewer (if applicable):
- Approved system of record:
- AI-tool/data-handling approval reference and allowed data class:
- Approved evidence storage, retention/disposition owner, and retrieval path:
- Active-change/concurrency check owner and result:
- Named implementation, validation, rollback, and incident-escalation authorities:
- Maintenance-window decision: `not requested / requested / approved`
- This worksheet is: `training / draft evidence / approved change record`

## 2. Evidence inventory

| Item | Source/version | Observed or proposed? | Owner verified? | Sensitive? |
|---|---|---|---|---|
| Business reason |  |  |  |  |
| Desired outcome |  |  |  |  |
| Affected service boundary |  |  |  |  |
| Constraints/dependencies |  |  |  |  |
| Baseline evidence |  |  |  |  |
| Validation evidence |  |  |  |  |
| Rollback prerequisite |  |  |  |  |

## 3. Human decision gates

- Does the packet include a specific approved change and owner? If no, stop.
- Is every source permitted for this tool and audience? If no, stop.
- Is the AI-tool/data-handling approval, evidence-storage location, and retention/disposition owner recorded? If no, stop.
- Is the desired state distinct from the observed state? If no, stop.
- Does a named human own the validation decision and rollback decision? If no, stop.
- Has the named owner recorded the active-change/concurrency check and a separately authorized implementation authority? If no, stop.
- Does the organization require a CAB, customer, security, or vendor review? If yes, obtain it outside this worksheet.

## 4. Bounded AI prompt

```text
Act as a documentation clerk, not a network architect, operator, security approver, change manager, or incident commander. Using only approved, non-sensitive facts supplied below, create: (1) an evidence inventory; (2) declared desired state versus observed state; (3) assumptions and missing information; (4) neutral reviewer questions; (5) proposed validation evidence; and (6) rollback prerequisites and stop conditions. Do not invent topology, commands, configurations, access, device behavior, capacity, security status, maintenance windows, or results. Do not recommend execution. Mark every unsupported item UNVERIFIED.
```

## 5. Handoff record

- AI output version and date:
- Input data class and AI-tool approval reference:
- Evidence storage location and retention/disposition instruction:
- Active-change/concurrency check result:
- Human review outcome: `not reviewed / revise / approved for next review only`
- Unresolved items:
- Exact action still blocked:
- Next named human:

## Synthetic drill

Pretend a team wants to improve resilience for a fictional office network but has supplied only: “Users report intermittent video calls,” a proposed maintenance week, and no approved topology, baseline, owner, or rollback record. A correct desk output is a gap list and stop condition—not a configuration, design, or change recommendation.