Hi @fresh-football-47124 and @helpful-application-7107 I have a experiment running with 100% traffic allocated to it. I want to run another experiment in parallel to the current but i want to avoid interactions between different experiment. I understand that i can use namespace but do you think enabling namespace for the another experiment that i want to run would avoid any interaction

Namespaces only work if you assign less than 100% traffic. There's no way to create a namespace that will only bucket new traffic between the two experiments. You'll have to restart and rebucket all users in both experiments with less traffic and both in a namespace if you want to run them in parallel.

@helpful-application-7107 how can we run holdout experiment in growthbook?

We don't have explicit support for holdouts but are building the framework necessary to support them. Can I ask what your use case is?

@helpful-application-7107 we are trying to run experiments on some group of users which are already divided into two groups (control and treatment) . We do not want Growthbook to do randomisation for us. We want to use GB for metrics estimations. Does GB supports such experiment?

Yes, you can use that, you just have to define the timestamps, the ids, and the variation id in a specific

Experiment Assignment Query

in your datasource to pull that list of IDs and the relevant timestamp from your data warehouse.

@helpful-application-7107 Hi, I found this thread searching for info about holdout groups! Just curious if there's a timeline on the shipping the framework that you mentioned. We're looking to create a long term holdout group to exclude from multiple experiments, and I was exploring namespaces as one possible solution, but doesn't look like that will work.

I don't think it's likely to land until closer to the end of the year/early next year. We have more and more of the back-end in place for them but aren't ready to roll it out yet. There is a pretty manual way to get close to something like an overall holdout group that I was just sketching out, but it hasn't been tested out before and it requires a lot of manual work so there's lots of room for user error (all stuff our eventual holdout feature would manage for you).

Got it, thanks! It's helpful to have that general timeline.