Hi team, I'm reading up on the Holdout docs and trying to figure out if we might be able to effectively use them in our setup or not. We are using url redirect experiments in a cloudflare worker at edge since our site is heavily cached so we need to check experiments for pages and do the bucketing and internal redirects before hitting our cache layer. We do not use the feature flag option because of the caching issue - the logic to check for a feature and show A or B in our server-side components would not be effective once the page is cached. Here are my queries re holdouts: • Is it correct that if you create a holdout group which then gets added to your experiments and a small percentage is bucketed as the holdout group, do you then need code logic (flags) to look for users in the holdout group to determine what to show them? If so - can you help me find docs/info on that? I can't find the info ◦ Or is it simpler and the holdout users simply see control and no code changes are required? • If we have config in our cloudflare setup to only run the cloudflare worker on pages with active experiments, will this be problematic if trying to use holdout groups effectively? • Could an alternative solution for us be to create new experiments as our 'post rollout' experiments designed to measure the effect of a winning change where we show/roll out the winning change to 90/95% of users and the remaining 5% (a holdout group) get the original, so that we can confirm in the following month or two that any positive impact is indeed from the rolled out change and not from other factors like a time of year that sees lots of traffic/growth? Any advice/info appreciated 🙏

@helpful-application-7107 do you know the answer?

Hi there Katrina. Let me discuss this with the team; we don't have an entirely clear flow here. Sorry about that.

Thanks @helpful-application-7107, will wait to hear back. We are open to using feature flags if we can find a way around our caching issue. I did see the docs say that feature flags can be used in an edge worker setup but guessing that would be pretty complex. I think for holdouts we're a bit blocked by the fact that in our current setup we're really only able to run experiments for new features/UI at a page level. Then if we roll it out the feature might show up on many pages. And holdouts are designed to help you hold back a small group from that wider rollout. Which sounds tricky with our page-level redirect experiment at edge setup. But any advice on potential solutions would be much appreciated! 🙏