Reviewing network config changes with Iris drift narration
A diff tells you which lines changed. It doesn’t tell you what the change means, which is what a reviewer needs. Iris’s drift narration turns a config change in rConfig into a structured explanation: what changed, what it affects, and how risky it looks. It’s the fastest way to triage a change you didn’t expect.
What it reads
| Input | Limit |
|---|---|
| Device name, what triggered the change, its severity | |
| The diff, secrets obfuscated | Up to 8,000 characters |
| The start of the current config, for context | The first 40 lines |
| The full previous and current configs | Only when no diff is stored |
The model is told this is a read-only analysis and not to suggest pushing changes.
How to read the output
Every narration has the same five sections, so reviewers learn where to look:
| Section | Read it for | Watch for |
|---|---|---|
| Summary | The gist in two sentences | A summary that doesn’t match the diff size: a one-line summary of a 300-line diff means detail was lost |
| Changes Detected | Each change grouped and explained | Changes to security-relevant lines (ACLs, AAA, SNMP, management access) |
| Operational Impact | What traffic, services or users are affected | Impact claims about things outside the device; the model can’t see neighbours |
| Risk Assessment | Triage priority | Treat it as a signal, not a verdict |
| Recommendations | What to verify next | Checks you can run now: interface state, routing table, a ping |
Where it fits in change review
- Unexpected change overnight: open the change in rConfig, click Analyze with AI, and decide in a minute whether it needs a call.
- Post-change review: narrate each device’s change and paste the narration into the change ticket as a plain-English record.
- Incident timeline: copy narrations for changes near the incident start into the timeline.
For changes you want explained automatically, without anyone opening rConfig, use event-driven narration. To check a change against its approved plan, use change window verification.
Limits to know
- Long diffs are cut at 8,000 characters. For a large change, the narration covers the first part. Review the rest in the diff view, or narrate a smaller change set.
- Context is the start of the config. The model sees the first 40 lines, not the whole file, so it may not know how a changed ACL is used elsewhere. Use config analysis on the current config for that.
- Not stored. Copy anything you want to keep.
Related reference
Frequently asked questions
What does Iris drift narration tell me about a change?
A summary, each change explained, the operational impact, a risk assessment and recommended checks, always in that order. It's generated from the diff and the start of the current config.
Can I trust the risk rating?
As a triage signal, yes; as a decision, no. The model sees the diff and some context, not your topology or business calendar. Use it to decide what to look at first.
Is the narration saved?
No. Narrations aren't stored. Use the copy button to paste one into a ticket or incident timeline.