Red flags and labels

​

What are red flags and labels?

A red flag is a tag Tuner attaches to a call when it matches a rule you define. Use red flags for failures that need attention, such as a failed hallucination eval or a severe latency spike. A label is created the same way and categorizes calls without marking them as failures, for example long_agent_monologue or sluggish_conversation. Both appear in Call Logs, and red flags also appear on the Overview dashboard. A rule has one or more conditions. With several, it tags the call when all of them match or when any one matches. To combine AND and OR in one rule, create it through the API with nested groups.

​

Create a rule

Rules also run on calls whose outcome has Stop Analysis turned on, so a rule like Call Outcome = Voicemail works. On those calls, conditions on Evals, Data Fields and Call Intent don’t match, including Not equal (!=).

​

Condition sources

The field list in each condition groups everything a rule can test into four sources:

​

Call metadata fields

How rules compare Call Metadata values:

• For text values, Equals (=), Not equal (!=), Contains and Starts with ignore case. For example, refund matches an outcome named Refund, and Contains hangup matches Customer_Hangup.
• Not equal (!=) on Call Outcome, Call Intent, Call Outcome Type or Call Ended Reason also matches a call with no value, for example a call on an agent with no outcomes set up. On Call Intent, a call whose outcome has Stop Analysis on doesn’t match.
• A call with no duration or cost never matches a Call Duration or Call Cost rule.

​

Example rules

Each example starts as one rule with one condition.

​

Hallucination detected

Any hallucination is a trust issue, so flag it whatever the call’s outcome.

​

Missed disclosure

To flag several compliance evals under one tag, add a condition for each one and choose ANY (OR). For example, add conditions for Identity verification compliance and Prohibited content.

​

Conversation freeze

A P90 latency above 3,000 ms is likely to feel like a hang or freeze to the caller.

​

Sluggish conversation (label)

Not every slow call is a failure. A label lets you filter and analyze the pattern without flagging it.

​

Long call that fails identity verification

This rule has two conditions. With ALL (AND), it flags only calls longer than 5 minutes that also fail the eval.

​

Where red flags and labels appear

​

Overview dashboard

Red flags appear in the Recent Red Flags section of the Overview dashboard, a queue of calls that need attention. Click View on a row to open the call.

​

Call Logs

Labels and red flags appear as color-coded tags on each call row in Call Logs, and you can filter calls by them.

​

Triage red flags

​

Investigate a red flag

Clicking View in Recent Red Flags, or a flagged call in Call Logs, opens the call details page. The recording and transcript sit on the left, so you can read what was said at the point of failure. The tabs on the right are:

• Overview: the call’s red flags and labels, sentiment and metadata.
• Evals: every eval result for the call, including the eval that triggered the flag.
• Voice Metrics: latency and other voice metrics for the call.
• Data Extraction: the values extracted from the call.
• Traces: the call’s trace spans, shown only when the call has a trace. This tab takes the full page width.

For a walkthrough of one call end to end, see Perform a root cause analysis on a single call.

​

Investigate patterns across calls

When the same flag recurs across many calls, use the Tuner MCP Diagnose workflow to look at those calls together and find the shared root cause.

​

Get alerted on red flags

Red flags don’t notify anyone on their own. To get notified, create an alert in Agent Settings > Alerts. An alert can use a specific red flag or Red Flag Count as a condition, and can combine several conditions with Match:. See Real-time alerts for the full walkthrough.

​

Manage rules

Each row in the Labels & Red Flags table has these actions:

• Edit: click the edit icon, change the rule, then click Update.
• Pause or resume: use the switch in the STATUS column. A paused rule doesn’t run on new calls.
• Delete: click the delete icon and confirm. Alerts that use the rule’s red flag pause until you update their conditions, unless another active rule applies the same red flag.

A rule with nested groups of conditions, which only the API can create, opens with its conditions read-only. You can still edit its tag and description, and saving keeps the conditions as they are. Rules follow the same nesting limits as alerts.

​

Resume a paused rule

Deleting an eval, data extraction field or voice metric pauses every rule that uses it in any condition. The rule’s row shows the reason, for example Eval deleted. Edit to resume., and has no switch. A red flag rule paused by a deletion can also pause the alerts on its red flag. To resume the rule, click Edit rule to resume, change or remove each condition that uses the deleted item and click Save and resume. If a condition still uses a deleted item or a voice metric that isn’t enabled, the rule stays paused. Otherwise it’s active again.

​

Limits

Limits enforced by the Add Rule dialog: To manage rules through the API, including rules with nested groups of conditions, see the Agent Settings group in the API Reference, starting with Create Rule and List Rule Conditions.

​

Next steps