Glossary

Guardrail metric

A secondary metric watched during an experiment to catch harm the primary metric would not show, such as load time or churn.

Also called: guardrail, health metric

A guardrail metric is tracked alongside an experiment's primary success metric to make sure a treatment that wins on the metric being optimized is not quietly causing damage elsewhere. Typical guardrails include page load time, error rates, unsubscribe or churn rate, and revenue, monitored even when the experiment is really about something narrower like click-through rate.

Guardrails differ from the overall evaluation criterion in purpose rather than mechanics: the OEC is what a team is trying to improve, while a guardrail is what it refuses to let get worse, and a statistically significant guardrail regression is usually treated as a reason to stop a launch even if the primary metric improved. Some guardrails carry formal thresholds ("must not regress load time by more than X"), others are watched more qualitatively.

Guardrails matter because optimizing a single metric in isolation reliably finds ways to move it that harm the broader product or business, a faster checkout that skips a fraud check, a more "engaging" feed that increases outrage. The common failure mode is defining too few guardrails and missing a real harm, or defining so many that some register a false positive by chance, which argues for correcting guardrail thresholds the same way a false discovery rate correction handles multiple comparisons.

Last reviewed September 22, 2026

In the index now

Related terms

Overall evaluation criterion (OEC) The single metric or composite formula a team agrees in advance will determine whether an experiment's treatment is judged a success. North star metric The single metric a team chooses as the best proxy for the core value a product delivers to customers. A/B testing A controlled experiment that compares two variants to measure which one produces a better outcome. KPI (key performance indicator) A measurable value tracked over time to show whether a team or process is on track toward a goal. False discovery rate (FDR) The expected share of "significant" results that are actually false positives, when many statistical tests are run and checked at once.

Related guides

How to build a dashboard people use Design a dashboard around its audience and the decision it supports, keep it maintained, and retire it when nobody opens it anymore. How to define a north star metric What makes a north star metric useful rather than decorative, the input metrics and guardrails around it, and examples by business model. How to read an A/B test result Significance, power, confidence intervals, sample ratio mismatch and peeking — what a dashboard's green checkmark does and does not tell you. How to set up an experimentation program From your first A/B test to a durable program — the plumbing, the statistics rules, the review ritual, and the tools that support each stage.