We are evaluating GB and so far I’m digging it. As I’m looking at it though, I’m wondering what is the most idiomatic way to persist a user their originally assigned experiment state, even after the experiment is turned off. e.g. I run a feature A/B. User is assigned Variant B. We then stop the experiement. I would like the user to remain in Variant B, even if B isn’t decreed the “winner” It sounds like what I want are perhaps “sticky” values and it sounds like that is in the works. But in the mean time, what are other folks doing to solve this problem?

@blue-sundown-32899 has marked the issue as closed. Reopen Issue button

(We are a React Native app)

First class sticky bucketing is in development at the moment, though we don’t have an exact ETA. Meanwhile you could probably rig something up with cookies and forcing a variation if the cookie is present

OK Thanks!. Do you have a general ETA for Sticky Buckets? (So I can gauge how much to invest in a homebrew solution).

Probably 1-2 months I'd think, but hard to say as we're still pretty early

super. Thanks!

Hey @happy-autumn-40938, we’re also interested in this feature. I guess you’ll make this optional per experiment, right? So we can decide per experiment if we want to have sticky bucketing or not.

As the dev primarily working on this, I'll say "very soon". Probably on the order of 1-2 weeks. We're not planning on a per-experiment on/off switch, although we will give you quite a few per-experiment controls on how you'd like to treat bucketed users after experiment changes (ie: re-randomize bucketing, persist existing bucketing, exclude bucketed users from the experiment, etc...).

Thanks! Not sure if it’ll help us though. We’re looking for an option that allows us to start an experiment with users in the US and, if they move to another country, keeps them in the experiment even if they no longer meet all the conditions. Does your proposed solution cover this?

Good call out. I've made sure bucketing persists against attribute-based targeting changes (such as US -> Canada for already-bucketed users)

Cool, thanks Bryce! Just to clarify, this would be something optional right? I mean, for some experiments we’d like to have this enabled, but for some of them we’d prefer having always the attribute-based filtering, even if they are already bucketed users.

The way we've planned it, you could achieve something like that a few ways: 1. Either force invalidate existing user buckets upon experiment rule change, or simply disable sticky bucketing for the experiment so that attribute based targeting directly gates bucketing. 2. Or keep sticky bucketing enabled and move your gating attribute rule onto the feature flag. Move it above the experiment rule so that it acts as a gate before any sticky buckets are evaluated.