Open source feature flagging and experimentation platform for stress free deployments, measured impact, and smarter decisions.

GrowthBook Users

They will stay in your experimental sample for analysis so long as they saw the experiment at least once (e.g. they are still counted as part of the dashboard results).

However, our feature evaluation is always done on the fly. If a user returns to an experiment check in your code and their attributes no longer match the targeting attributes, then they don't match the targeting and won't get the experiment value!

To prevent this behaviour, you need to enable sticky bucketing: <https://docs.growthbook.io/app/sticky-bucketing|https://docs.growthbook.io/app/sticky-bucketing>

How does this interact with the choice of whether to re-randomize traffic in a new phase?

You can make explicit choices about what happens to users with sticky bucketing. Our "Make Changes" UI does the best to guide you to understand the trade-offs. There's also a bit more written here: <https://docs.growthbook.io/app/experiment-configuration#making-changes-while-running>