Fine-grained control over releases is crucial for modern software delivery teams, but without change monitoring, that control becomes futile. LaunchDarkly Guarded Releases allow teams to perform a guarded rollout, where new features or changes are gradually rolled out to a portion of users, with real-time monitoring and automatic reversion if key metrics indicate a problem. Understanding regression thresholds is essential for fine-tuning risk tolerance in guarded rollouts, as it sets the level of underperformance that can be tolerated before rolling back the release. A lower threshold means being more cautious about performance drops, while a higher threshold allows for slightly worse performance within a set buffer zone. Regression thresholds are relative concepts, not absolute values, and must be measured against the original variation's performance rather than a fixed percentage. Setting a regression threshold correctly involves understanding how to define "worse" in the context of the success criterion specified in the metric definition, which can depend on whether higher or lower is better. Ultimately, a regression threshold defines the additional level of underperformance that can be tolerated before rolling back the release, and setting it correctly requires a nuanced understanding of relative differences from original variation performance.