Manifesto
Restored is not fixed.
7 principles for network teams who want proof, not relief. Share them, argue with them, and apply them with or without us.
Restored is not the same as fixed.
Completed is not the same as verified.
Configured is not the same as compliant.
Why
Relief is not proof
Most network incidents are closed when service comes back, not when the cause is known. Most changes are marked successful because nobody complained. Both leave risk in the network that surfaces later, at a worse time, in front of whoever is on call.
The answer is not another dashboard. It is a habit of evidence.
Restored ≠ Fixed: REAP in 90 seconds
Read the video as text
0:00–0:08 21:14. A change goes in. Closed “successful.” CHG-1234 edits object group OG-ERP on plant-a-core-2. Closed at 21:20. No post-check.
0:08–0:16 21:16. The ERP path fails. Nothing alarms. ACL-ERP-IN matched OG-ERP. ERP return traffic now hits the final deny. Link and OSPF stay up.
0:16–0:24 02:07. P1. Plant A can’t enter orders. The night-shift help desk opens INC0451. Undetected for 4 h 51 min.
0:24–0:33 Every incident starts from zero. Six tools. Three teams. The answer is in there, spread across all of them. Engineers spend 37% of their time addressing disruptions.
0:33–0:42 REAP observes. Read-only. Config history, ACL counters, routing and the change ticket, gathered into one investigation.
0:42–0:52 It reasons with evidence, one step at a time. What changed, why it broke, what is affected. Each line traces to its source.
0:52–1:00 It recommends an approved runbook. A person decides. Restore OG-ERP to its 21:13 members with RB-07. Awaiting approval. Approval-gated runbooks: pilot deliverable today.
1:00–1:10 Before and after. A verdict, in the ticket. ERP path re-tested after RB-07: Verified. CHG-1234 recorded as Rolled back, with evidence. Before-and-after verification: pilot deliverable today · automated verdicts: in development.
1:10–1:18 Start through one door. Incident Reasoning. Change Assurance. Resilience Assurance. Expand only when you pull.
1:18–1:24 Restored ≠ Fixed. Completed is not the same as verified. Configured is not the same as compliant.
End card, 1:24–1:30 REAP · Observe · Reason · Act · Prove Agentic Network Assurance. Grounded in Evidence. Bring us your last P1 · reapnet.ai
Illustrative, on a fictional network. · Source: New Relic (opens in a new tab), 2026 Observability Forecast (opens in a new tab)
The principles
7 principles of network assurance
Every incident closes with a cause, not just a recovery.
A reboot or a failover restores service; it does not explain it. The incident is done when the root cause is written down with the evidence behind it and the fix is verified.
Every change closes with evidence.
State the intent, capture the state before, compare the state after and record one outcome: Verified, Degraded, Rolled back, Inconclusive or Blocked. “Completed” is a workflow state, not an outcome.
Redundancy is tested, not assumed.
A standby that stopped synchronizing protects nothing. A failover tested at commissioning is a memory, not a fact. Prove it works before the failure that needs it.
Conclusions show their evidence.
Every answer, from an engineer or an agent, links to the command output, log line, configuration diff or ticket text it rests on. When the evidence is thin, say so and name the next diagnostic step.
Agents earn autonomy.
An agent starts read-only. It earns the right to recommend, to verify and then to act with approval, one workflow at a time, on measured results. The dial stays in the operator’s hand.
Measure P95, not averages.
An average hides the investigation that took all night. Report the median and the 95th percentile of time to answer, and never hide the tail.
The ticket is the record.
Evidence belongs in the incident, problem and change records your team already uses. A finding that lives only in a chat thread or on a dashboard is lost at the next handoff.
Hear it, then test it
At ONUG in New York, then on your own network
Raja Tadimeti, REAP’s co-founder and CEO, presents “The Change Succeeded. Did the Network? Building Confidence in Change Outcomes” in the Tools, Technologies & Techniques theater at ONUG Fall 2026 (opens in a new tab) on Thursday, October 29 at 3:15 pm ET, at the PENN DISTRICT, 100 West 33rd Street, New York. It is a problem-first talk built on one realistic change that was marked complete and was not verified, with a validation checklist you can apply with or without AI. REAP is a sponsor and speaker; book 20 minutes with the team, or take the 8-question assurance check at the booth.
Then test the principles on your own network. Bring us your last P1: keep the root cause to yourself; in a 30-minute session we walk through how REAP would have investigated it, evidence first. The replay on your own network happens inside the proof of value.
Bring us your last P1.
Keep the root cause to yourself; in a 30-minute session we walk through how REAP would have investigated it, evidence first. The replay on your own network happens inside the proof of value.