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

# Observability

> Conversation logs, alerting, analytics, exports, SLA monitoring, surveys, validated deflection, and knowledge gap logging in Everest.

## Real-Time Conversation Logs

Everest shows live and historical transcripts with full turn-by-turn detail. Each turn exposes the tool calls made, captured variables, retrieval results, and any errors, so operators can see exactly what the agent did and why. Live sessions stream in real time and historical sessions are searchable.

## Error Visibility and Alerting

API, flow, and conversation failures surface directly in the transcript view and in an error dashboard. Alert rules fire on configurable error-rate thresholds and notify via email, Slack, or PagerDuty. Each alert links back to the affected conversations for immediate diagnosis.

## Conversation Analytics Dashboard

The built-in dashboard reports:

* Validated deflection rate and escalation rate
* CSAT score
* Intent distribution
* Top failure points
* Session volume

All metrics are available out of the box with no report configuration required.

## Custom Reporting and Data Export

Raw conversation data exports through scheduled jobs to Snowflake, Tableau, and Power BI connectors, or on demand via REST API in CSV or JSON format. Scheduled exports run on a configurable cadence, so downstream BI models stay current.

## Bot Health Monitoring and SLA

Everest carries a 99.9 percent uptime SLA. Latency is monitored per flow and per API integration, and incidents trigger notifications to configured channels. SLA performance is visible alongside the analytics dashboard.

## CSAT and Post-Interaction Surveys

Surveys deliver natively per channel: inline at the end of a chat session, as an IVR prompt at the end of a voice call, and as an email follow-up after escalation or case creation.

* Format is configurable per channel: thumbs, stars, NPS, or open text.
* Negative-feedback reason lists are configurable, so follow-up prompts match your taxonomy.
* Results filter by channel, contract tier, and issue type.

## Validated Deflection

A resolution counts as deflected only after an automated quality review confirms the answer was accurate, complete, and confirmed by the user, not simply because the user did not escalate. Judgments below the confidence threshold, and any disputed resolutions, are flagged into a human sampling queue for review. Confidence thresholds are configurable per deployment.

## Knowledge Gap Logging

Every unanswered or low-confidence query is logged automatically with the channel, the user's input, the knowledge articles that matched, and the retrieval confidence score. A gap report groups recurring gaps so the content team can prioritize what to write. When a gap is closed, the new article links back to the originating gap record, completing the workflow from detection to KB update.