# Read an on-call investigation

Understand a night incident in a few minutes, without opening the consoles one by one.

## What happens without you

When a node or a cluster goes down, the incident is confirmed, the subscribed people are notified, and the root cause investigation starts on its own: hypervisor logs, recent changes, dependencies. The alert arrives with a first analysis; the full analysis follows when the node comes back.

## The steps

1. Open **Monitoring › Incidents**. The night's incident is there, with its duration and state.
2. In the **Root cause** column, click **Analysis**.
3. Read the summary, then what is **established**: each fact cites its evidence, a log line or an event.
4. Read the **hypotheses**. They are competing readings, not a verdict: each gives its confidence, what supports it, what contradicts it, and the check that would settle it.
5. If a next step is suggested, **Apply** opens a review of the actions: nothing runs before you confirm, and each action goes through the same locks as in the inventory.

## Ask the Incidents agent

Click the **Incidents** agent at the top of the console, then **Discuss**: ask about the impact on its scope, or what to watch next. It reads the investigation rather than redoing it.

### The investigation does not say who did what

The collector cannot read this source's logs yet, or the collection account lacks the right. The investigation page says so before it even starts, and the Incidents page offers to update the collector.

### No hypothesis is proposed

That is a result, not a failure: no cause is supported by the available elements, and the investigation refuses to invent one.
