Changelog: Multi-armed bandits, Launch Insights project comparison, and more
Multi-armed bandits
Multi-Armed Bandits (MAB) are now available to all Experimentation customers! This adaptive testing method automatically reassigns your audience toward better-performing variations as results emerge, helping your teams learn faster, reduce opportunity costs, and get the best experiences into their customers’ hands sooner.
MABs work seamlessly with our redesigned Experiment Builder, updated results screens, and our latest metrics enhancements. It’s a major milestone in our journey to deliver powerful, flexible experimentation at scale.
To create your own Multi-armed bandit, just follow these steps.
Project comparison for Launch Insights
Launch Insights just got a major upgrade: Teams can now compare performance across up to four projects side-by-side. This feature helps organizations understand trends across teams, track adoption, and identify outliers more easily.
In addition to comparison functionality, we’ve introduced improved filters, a more intuitive date picker, and a refreshed chart design that boosts clarity while reducing noise. We’ve also adjusted how we display scores in comparative views to better reflect context, especially for newer accounts. To learn more, read Launch Insights.
Other improvements
Lots of improvements to the new individual context targeting experience based on customer feedback
Added stage duration and rollout percentage details for approval requests
Improved metric regression display by sorting results chronologically
Added frontend validation for maximum 10 phases in release pipelines
Redesigned the welcome experience to guide new users more efficiently with role-based onboarding
Various experimentation improvements, including updated tooltips for Bayesian experiments, funnel experiments only requiring a primary metric, and more
Users can now filter error distributions by date
Dashboard key management now includes visible keys, real-time duplicate detection, and better error messages
A new topic in LaunchDarkly Docs now outlines tips and best practices for setting up your context kinds to run effective guarded releases
Bug fixes
Fixed percentile metrics chart display inconsistencies where chart values didn't match tooltip values
Resolved event selector disabled state issues in composite event details views
Fixed layout problems on various list view screens throughout the application
Corrected funnel step tooltips that were incorrectly showing decimal values
Fixed trial banner display logic to properly handle different account states and environment