SDK Concepts

SDK concepts are the shared pieces every A vs B SDK is built on. They cover how it knows who a visitor is, how it splits traffic, and how it logs what it did. Each page below tells you what the concept is, why it matters, a real example, and when to skip it.

New to A vs B? Start with Getting started instead. Come back here once you need one of these.

Start with multi-context identity if you target more than one kind of thing, such as a user and their company account. From there, learn what's built on top of it. A holdout is a slice of visitors kept out of every experiment. It gives you a clean baseline to compare against. Sticky bucketing means keeping a visitor in the same variation (the version they were tested with), even after you change the traffic split.

Multi-context identityHow kinded contexts let you target a flag by user, organization, device, or any entity you define, in one call.

HoldoutsHow a holdout group excludes the same visitors from every experiment at once, for a clean baseline.

Sticky bucketingHow sticky bucketing keeps a visitor's variation stable, even if you change the traffic split partway through.

BanditsHow a bandit rule shifts traffic toward the best variation on its own, using epsilon-greedy, Thompson sampling, or UCB1.

Decision loggingHow to log every flag decision and send it to your own warehouse or observability tool, like Snowflake or Datadog.

OpenFeature compatibilityHow A vs B's OpenFeature providers work for server and browser SDKs, and when to pick them over the native SDK.

CleanupRegistryHow CleanupRegistry stops resource leaks, like open timers and connections, when you close the SDK client.

Typed contextsHow defineContextSchema() checks your targeting attributes so a typo gets caught before it reaches production.