> ## Documentation Index
> Fetch the complete documentation index at: https://docs.resolve.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Escalation Decisions and Criteria

> How ResolveAI decides to close or escalate a page, and how a team shapes those decisions with escalation criteria and feedback.

Every page ResolveAI handles ends in an escalation decision. Escalation criteria are the team's rules for that decision, and feedback is how the team corrects it.

## Escalation decisions

A decision has one of two outcomes.

| Outcome | What happens |
| - | - |
| **Closed** | Resolve closes the page. The on-call engineer is not paged. |
| **Escalated** | Resolve escalates the page. The on-call engineer is paged straight away, with the reason and the evidence behind the decision. |

During [trust building](https://docs.resolve.ai/primary-on-call#trust-building), the same decisions are shown as **Would close** and **Would escalate**. Resolve records the decision but changes nothing in the paging tool.

### What a decision shows

Open a decision from the alert thread or the **On-call & Investigations** page to see:

* **The reason**, in plain language, under **Why it closed** or **Why it escalated**.
* **Evidence**: the observations the decision rests on, each with a citation.
* **Timing**: when Resolve decided, and how long after the alert fired.

## Escalation criteria

Escalation criteria are plain-language rules that define what is worth paging a person for. Resolve reads them on every page it decides.

### Find the criteria

1. Open the team, then go to **Settings** > **On-call and Investigations**.
2. Select the **Escalation decision criteria** tab.

The tab appears once the team has a paging tool connected. Org admins can edit the criteria for any team. Team admins can edit the criteria for the teams they administer.

### Edit the criteria

Each team starts with default criteria, which are generic. Edit them to match what the team considers worth a page.

1. Edit the text on the **Escalation decision criteria** tab. The criteria are free text, up to 40,000 characters.
2. Click **Save rotation criteria** the first time, or **Save new version** after that.

Resolve uses the new criteria on the next page. Every save creates a new revision. Click **History** to review earlier revisions, see who saved each one, and **Restore** any of them.

### Write good criteria

* **Keep the criteria high-level.** The criteria apply to every alert the team's filters cover, so write rules that hold for all of them.
* **Put alert-specific guidance in runbooks.** Resolve also reads the runbook attached to an alert when it makes the escalation decision, so guidance for one alert belongs there. See the [Knowledge Setup Guide](https://docs.resolve.ai/knowledge-setup-guide.md).
* **Name the impact that matters.** State which services, customers, or user journeys justify a page at any hour: "Escalate any alert that shows failed customer requests on checkout or login."
* **Name what can wait.** Describe the kinds of pages that are safe to close: "Close alerts that recovered on their own with no customer impact."
* **Say what to do when evidence is missing.** If a dashboard or log source is unavailable, say whether that alone is a reason to escalate.
* **Keep it short.** A short list of clear rules works better than a long document.

## Give feedback

Feedback is recorded against the exact decision and the criteria revision it used. Leave it wherever the decision is being read.

> [tab] Resolve UI
>
> On the decision, click thumbs up or thumbs down, then answer:
>
> * **Was Resolve's response correct?** Choose **Looks right** or **Needs work**.
> * **Should Resolve have paged a human?** Choose **Yes** or **No**.
> * Add a comment to say what Resolve should have done differently.

> [tab] Slack
>
> On the decision in the alert thread, click **Right call** or **Wrong call**. The feedback form asks the same questions as the Resolve UI.
>
> To turn the feedback into a criteria change, select **Propose a decision-criteria change** in the form. Resolve replies in a separate thread with the change it would make. Click **Approve** to save it as a new revision, **Adjust** to ask for a different change, or **Reject** to leave the criteria as they are. Approving needs the same role as editing the criteria.

> [tab] MS Teams
>
> Open the feedback form on the alert thread and answer **Was Resolve's response correct?** and **Should Resolve have paged a human?**

Feedback alone does not change the criteria. The criteria change only when someone edits them or approves a proposed change.

## Tune the escalation rate

Expect a high escalation rate at first. The default criteria are general, and escalating is the safe default when Resolve is unsure.

To bring the rate to the level the team expects:

1. Review decisions during trust building, starting with every **Would close** decision. Confirm that each one is correct. Resolve always escalates when it is not certain, so a close is the decision that must not be wrong, and these are the most important cases to check.
2. Review the **Would escalate** decisions the team disagrees with.
3. Leave feedback on each decision reviewed. For a wrong **Would close**, say why a person was needed. For an unneeded **Would escalate**, say what made the page safe to close.
4. Add the patterns that come up repeatedly to the escalation criteria.
5. Check the next few days of decisions, and repeat.
