Guardrail metrics

Overview

This topic explains how to use guardrail metrics to provide a consistent set of default metrics for evaluating guarded rollouts and experiments.

Guardrail metrics can help organizations standardize how they measure the safety and success of feature releases. By defining a trusted set of metrics in a dedicated metric group, admins can ensure that every guarded rollout and experiment starts with a consistent baseline for measurement. This reduces setup time, promotes standard practices, and helps teams move faster with confidence.

Prerequisites

To create or update guardrail metrics, members must have one of the following:

• an Admin or Owner base role in your LaunchDarkly account, or a role with guardrail metrics group permissions
• At least one custom metric defined in the project
• A consistent randomization unit, such as user or account, across the metrics you want to include

How guardrail metrics work

To begin, Admins, Owners, or other members with proper permissions define guardrail metrics for a project. Then, when any member creates a new guarded rollout or experiment, LaunchDarkly displays “Guardrail metrics” at the top of the metric group selector and includes them automatically.

This ensures that each release starts with a consistent set of trusted metrics. Teams can still customize their configuration, but the default behavior encourages alignment across teams.

Auto-attachment behavior

When you create a new guarded rollout, LaunchDarkly handles guardrail metrics as follows:

• LaunchDarkly checks whether the project has a guardrail metrics group
• If the group exists and its metrics use the same randomization unit as the guarded rollout, LaunchDarkly automatically attaches the guardrail metrics group
• If no compatible randomization units exist between the rollout and the guardrail metrics group, LaunchDarkly does not attach guardrail metrics

LaunchDarkly does not automatically add guardrail metrics groups to experiments. You can manually add them as secondary metrics if they match the experiment’s randomization unit.

You can manually remove a guardrail metrics group from a rollout or experiment if needed.

Limitations

Guardrail metrics groups include the following limitations:

• Each project can define only one guardrail metrics group
• All metrics in the group must use the same randomization unit, such as user or account
• Guardrail metrics apply across an entire project and do not vary by team, application, or SDK

Admins can manage the group from the guardrail metrics page in the Metrics section.

Example

An engineering leader creates a guardrail metrics group that includes API latency and system error rate, two key indicators of reliability. When a team sets up a new guarded rollout, LaunchDarkly includes these metrics automatically. This ensures that every release is measured against system health standards without requiring additional setup or guidance.