Flags & experiments
Ship behind a flag, then prove it worked
Feature flags with per-environment configuration and local evaluation, and experiments with sequential statistics that stay correct however often you look.
Available
Flags
Boolean, multivariate and remote-config flags with ordered targeting rules. Bucketing is bit-identical in every SDK, proven by shared test vectors. Server SDKs evaluate locally with no network call per check.
- Separate configuration for production, preview and development
- Users keep their variant when they sign in
- Going live in production needs a named scope, so an automation cannot flip it by accident
- Flags are free on every plan
Experiments
Sequential testing by default, with fixed-horizon and Bayesian analysis available. CUPED is on by default. A sample-ratio mismatch blocks the verdict instead of reporting a wrong one.
- Proportion, mean, sum and ratio metrics; guardrails as non-inferiority tests
- Holm and Benjamini–Hochberg corrections for many metrics
- A power calculator before you start
- The statistics engine is MIT licensed and checked against scipy, statsmodels and gbstats
Go deeper: Feature flags · Experiments