Hey, sorry if I've missed something obvious, but where can I manually set the initial weights for a bandit's variants?

With bandits, the initial allocation is equally distributed to allow for the exploratory stage. There isn't an option to weight the variations. Do you want to share more about your use case?

Paywall testing with UA through ads, set to a bid cap. Looking to limit downside risk to not give FB bad signal on eCVR @strong-mouse-55694 I could create multiple variants that go to the control on the frontend. Say 10 variants but 7 actually go to control. Would that be ridiculous?

Hi Henry, I'm a data scientist at GrowthBook. Could you use a short initial exploratory period, so the weights start quickly moving towards the winner? If you use 7 variants going to control, then it might be tough for the bandit to identify the correct variant, as the 7 controls will compete with each other. Luke

Hi Henry, I'm another Luke data scientist at GrowthBook 🙂 You could create a a/b test and use that as a Prerequisite for getting into the bandit to achieve what you're looking for. Your 7 out of 10 arms going to control idea is also fine, but it could end up being weird for a bunch of reasons, so splitting your experiment up into two stages seems to make sense to help you measure the overall efficacy of the bandit as well as reserve some traffic for the status quo. You could always, of course, run the bandit on a small percentage of traffic to achieve the same goal, but then you don't measure that "control" vs bandit effect.

Thank you so much, Lukes! This is super helpful