Managing experiments

The Experiments page shows every A/B test and feature experiment detected in your codebase.

What it detects

The scanner identifies experiment patterns from major providers:

• Optimizely — getExperiment(), activate()
• GrowthBook — useGrowthBookFeatureValue(), getExperimentValue()
• PostHog — experiment-related abTest(), splitTest()
• Statsig — checkGate() with experiment names
• LaunchDarkly — variation() with experiment-like flag names
• Split — Split SDK calls
• VWO — Visual Website Optimizer patterns
• Unleash — variant-based experiments

For each experiment, the scanner extracts: name, provider, variants (from arrays or if/else branches), hypothesis (from nearby comments), metric, and lifecycle status.

Experiment statuses

Cleaning up concluded experiments

Concluded experiments are tech debt. The branching code remains in the codebase even after a winner was chosen. The Experiments page flags these with a warning:

"N concluded experiments still in the codebase. These should be cleaned up — ship the winner and remove the branching code."

To clean up:

1. Identify the winning variant from your experimentation platform
2. Remove the experiment check and the losing variant's code path
3. Keep only the winning behavior
4. Remove the experiment configuration from your provider

In the feature guide

Each scanned feature's guide shows its experiments in the "Experiments & A/B tests" section of the reference tables. The guide's deep dives also note which user flows branch on experiments and what each variant does differently.