07/25/2023, 10:03 AM
I don't quite understand why this isn't allowed? What if I want to have two in the "off" state and one on the "on" state? I have two feature keys using the same experiment key to coordinate behaviour in two places for the same experiment


07/25/2023, 11:19 AM
Hey James - I'm not sure I understand your use case 100%, but in regards to your question about why you can't have more than 2 variations - it looks like you've selected a
type flag - which inherently only provides two possible values -
. Selecting a different feature flag type (
, or
) does give you the option to add more than 2 variations. Given your use case, rather than creating two feature keys using the same experiment key to coordinate behaviour, is there a reason you couldn't check the same feature twice in the two different places?


07/25/2023, 1:48 PM
Not really possible in our use case
These are the two experiments
We often use fixed feature keys to control specific parts of our app - e.g. which onboarding flow (from our CMS) to use, what pricing behaviour to use, etc
so e.g. in this experiment, we're updating our "welcome" flow to orient around users making a commitment to sticking to the program
But to do that, we not only need an adjusted onboarding flow, but also to make a small adjustment to other behavior. To coordinate that, we just use the same experiment key on the relevant features
In this case, we have 3 variations - a control, a variant with one adjusted behavior, and one with two adjusted behaviors
I don't think there is actually any valid basis for denying the ability to use e.g. true/false more than once


07/25/2023, 4:06 PM
@busy-horse-73824, my guess is we designed it this way initially to avoid confusion for users, but I definitely see your point that allowing experimenters to have multiple groups with the same boolean flag makes sense. Of course, a work around is to use a different flag type for now. However, the longer term solution that we are working actively on having an experiment serve as the source-of-truth for variations. In this case, you could explicitly link multiple features to the same experiment, and in this process we'd need to handle the case where multiple variations got a
value for a boolean flag. (cc @future-teacher-7046)
👍 1


07/25/2023, 8:58 PM
Ah that does seem like a much better way around, in terms of data model. I currently do quite a lot of teaching around how to run more advanced/coordinated A/B tests like this one, but that would make it much simpler I think
I suspect a
type where I just type in `true`/`false` will resolve the issue for now