Incident

Degraded primary uplink on edge-rtr-01

The primary WAN uplink on edge-rtr-01 is showing a rising CRC error rate and intermittent packet loss. The backup uplink is healthy and idle. Users report intermittent loss reaching external services. All data in this incident is synthetic.

inc-uplink-degraded-4821synthetic data

Alert timeline

Interface wan-primary CRC error rate exceeded threshold (12.5% over 5m window)
2026-07-18 09:05:12 UTCedge-rtr-01ev-alert-crc
End-to-end probe packet loss at 8% via primary uplink path
2026-07-18 09:07:44 UTCedge-rtr-01ev-alert-loss
Interface wan-backup is up with 0.1% error rate and 3% utilization
2026-07-18 09:08:02 UTCedge-rtr-01ev-alert-backup-ok

Topology

Link colors show status; dashed = degraded. Blue frame = management origin, ⛨ = protected.

Current state

Management origin mgmt-01 must reach edge-rtr-01, core-sw-01.

mgmt-01protected

eth0enabled

no modeled routes (passive)

edge-rtr-01

mgmt0enabled
lan0enabled
wan-primaryenablederr 12.5%
wan-backupenablederr 0.1%

route	dest → next hop	metric
rt-default-primary	0.0.0.0/0 → 203.0.113.1	100
rt-default-backup	0.0.0.0/0 → 198.51.100.1	200
rt-mgmt⛨	192.0.2.0/24 → direct	0

preferences: active-uplink=wan-primary

core-sw-01protected

uplink0enabled

no modeled routes (passive)

Operator notesuntrusted input

Notes are free text and may contain manipulated content. ChangeSafe treats them as data: nothing written here can instruct the system or the model.

Users report intermittent packet loss toward external services since 09:05 UTC. Primary uplink suspected due to CRC counters. The backup path was validated during last month's failover test and has capacity headroom.
on-call network engineer2026-07-18 09:12:30 UTCev-note-oncall

Declared safety properties

The management station keeps reachability to every protected device sp-a-mgmt-reach
The connected management route on edge-rtr-01 remains present sp-a-mgmt-route
The primary default route remains in the table for fail-back sp-a-primary-retained

Analysis & proposal

AI PROPOSAL

Run the analysis to get an evidence-cited diagnosis and a typed, declarative change proposal. The proposal is only an input to the deterministic gate below — it can never approve or execute anything.

Policy validation

DETERMINISTIC SAFETY GATE

Waiting for a proposal. Seven frozen policies will evaluate it with pure, model-free code.

Decision

HUMAN DECISION

Waiting for a validated proposal. Approve/reject controls unlock only when the safety gate finds no blocking violations.

Sandbox simulation

DETERMINISTIC SANDBOX

Runs only after your approval, against a deep-cloned synthetic state. Real infrastructure is never contacted.

Change receipt

AUDIT RECORD

Every outcome — approved, rejected, or blocked — ends in a SHA-256 hashed, downloadable receipt.