configure-trace-metrics

You can turn OpenTelemetry spans into custom trace metrics by configuring numeric metrics from span attributes to track latency, token usage, and workflow signals. This skill sets up those metric definitions for production environments.

Yes, you can deploy trace-aware LLM judges that utilize both transcript and OpenTelemetry trace context. These LLM evaluation metrics verify tool grounding and recovery behavior within agent workflows.

You can create high-signal operational metrics like P90 latency, tool failure rates, and workflow completion rates from trace data. This automatically pairs new instrumentation with actionable performance monitoring.

Yes, you need Coval CLI access and existing OpenTelemetry trace data to validate metric definitions and aggregation methods. This environment is required to properly configure and test the trace-based metrics.

The best way to evaluate tool usage and latency is by configuring custom trace-based metrics from OpenTelemetry spans. This approach transforms raw, noisy trace data into meaningful business and operational insights.

Relying on raw trace data often provides no clear decision-making value because the data is noisy. Configuring trace-aware metrics and LLM judges is needed to extract meaningful operational and business signals from spans.