Most reports that "the platform is wrong" turn out to be configuration rather than faults. This article covers the investigation sequence and the eight symptoms that account for the majority of cases.
The investigation sequence
- Start at the incident, not the anomaly list. Open the incident and read severity, main cause, and subcategories.
- Check the window. Last Sync on the site, and the time-range picker on the view. Both change what you are seeing.
- Confirm the scope. Which sites, devices, and clients are affected — and which are not. What is fine narrows the cause fastest.
- Order the observations. Sort by start time. The earliest is the prime suspect; the rest are usually downstream of it.
- Look at what changed. The Changes Log for config, VLAN, and SSID changes; Assets for a firmware version that moved.
- Open Statistics from the observation. The observation line opens Statistics with the involved metrics and the correct time range already selected. Rebuilding that view by hand takes minutes and introduces mistakes.
- Read Insights and follow the chain. Then test it: what would be true if this were the cause, and is it?
- Apply Recommendations, then tune. Fix it, then change a setting so the same ticket does not return next month.
Steps 2 and 5 are the ones people skip under pressure, and they are the two that most often change the answer.
Start with sync status
Every site view carries a Last Sync time and result, and it tells you what window the analysis is working from. A site that has not synced — or a report generated days ago — will confirm almost any hypothesis you bring to it, including a correct one for the wrong reason.
If a site shows no devices, that is an integration or sync question rather than a network one, and the Intents Log will say which.
Eight symptoms and where they come from
| Symptom | Usual cause | Where to look |
|---|---|---|
| A site shows no devices | Integration or sync, not an outage | Last Sync and the Intents Log. For Meraki, confirm the token admin belongs to that organization. |
| Availability looks better than the phone calls | Devices missing from the count | Assets versus what exists on site, then Site Management for misattribution. |
| Utilisation never alerts | Bandwidth limit not configured | Settings › Network Interfaces. Without a limit the metric is statistics only. |
| The QoE score looks wrong | SLA profile not tagged | Settings › QoE. An untagged profile is scored against the wrong expectations. |
| A switch site is drowning in alerts | Port roles left at defaults | Settings › Network Interfaces. Mark client ports; mark trunks as critical. |
| Severity changed on its own | Escalation timers, working correctly | Settings › Sites. Read the initial severity and escalation intervals for that device class. |
| People stopped reading the emails | Over-subscription | Integrations › Notifications. Remove anomalies and utilisation topics from external contacts. |
| Unrelated devices keep sharing a ticket | Device-to-site attribution | Settings › Site Management. Reassign, then watch whether correlation improves. |
Seven of those eight are configuration rather than faults, which is why the settings review at onboarding matters more than it looks.
Four ways an investigation goes wrong
- Stopping at the classification. The main cause names the incident correctly, but it cannot see the unscheduled backup job, the contractor near the fibre, or the unannounced ISP maintenance. Ask what happened outside the network that produced this pattern.
- Stopping at the first fit. The first explanation that fits gets adopted because it is a relief. Ask which observation actually came first, and whether it supports the theory.
- Reasoning off a stale view. Ask when the site last synced, and whether anything is missing from the totals.
- Treating the fix as the cause. "We rebooted it and it came back" is a mitigation, not an explanation. Ask why it needed that intervention in the first place, and read the Recommendations before closing.
Where the evidence lives
Observations hold the unfiltered record inside the incident. Statistics holds every historical metric across the site's devices and clients, down to five-minute granularity. The Changes Log holds every device, VLAN, and SSID added, modified, or deleted. Network Topology shows what sits behind what, and the VPN Graph shows link status and live traffic across interconnected sites. The Device Dashboard carries CPU, memory, sessions, and status for one device, with its own incidents and anomalies.
Related articles
- Reducing alert noise
- Anomalies, observations, and incidents
Comments
0 comments
Please sign in to leave a comment.