We have an A/B/C test and we want to kill one of the groups and keep running the experiment with only two groups. Should we assign 0% of traffic to C to accomplish this? Is there a way to reassign users in group C to either A or B? If we do a new A/B experiment with a different tracking key we are at risk of users that got B before getting A now, right?

Hi team, Can we access Growthbook's service via VPN peering? Currently, I face some rate limiting issues and was thinking VPC peering may reduce latency.

Hi all, I just got access to our company's growthBook earlier today and was playing around with it a bit to get to know the ins and outs and discovered something which I think is strange. Specifically when you delete a feature and then create a new feature with the same name. History and versioning seems a bit messed up. See below some steps to reproduce: 1. create a feature 2. make some changes resulting in 2 new versions to be created (version 2 and 3) 3. delete the feature 4. create a new feature with the same name expected behaviour (option a): • a new feature is created without any trace of the old feature's history expected behaviour (option b): • a new version of the old feature is created with the trace of the history and the current version is version 4 (or 5 if you consider the deletion being version 4) actual behaviour: • a feature is created with version number 1 (but versions 2 and 3 of the old feature are accessible, as well as the audit log) • if you create a new version, this will be version 2 (the draft will be shown next to the existing version 2), will this overwrite the original version 2, will there be 2 version 2's, or will the system crash, unsure (and I'm not intending to test this out 🙂) __ I doubt that this is expected behavior, but since I only played around with the tool for an hour or so, my knowledge in the tool is still limited. Could you please advice? Is this expected or actually a bug? Do I report this in github, or do you take over from here? Thanks, Brecht

Hi, we started to run into the issue

Please reduce your request rate. (Service: Amazon S3; Status Code: 503; Error Code: SlowDown;

when setting the

Experiment Auto-Update Frequency

for our experiments. We set the value to 48 hours and now that we have 5 running experiments (we expect this number to grow), we see query errors almost every time on the Auto-Update. Have other users reported similar issues? Do the Auto-Updates trigger all queries in all experiments in parallel? Or are there mechanisms in place to prevent request rate errors?

Similarly we want to be able to slowly decrease traffic to an experiment.

If I convert from a rollout rule to an experiment rule, will the groups of users be the same?

Hi all, I see that one of the data sources options is postgresql. I’m just starting out and will be fine creating a new db and putting data into it. Is there a guide as to what the schema should be and how I should connect GrowthBook to that new db?

3 questions: 1. I have one feature flag (it sets a coefficient to 0.5), ran an experiment yesterday , then archived it and marked it as "Lost". I proceeded to try a new experiment with the same feature flag but different variants (e.g. 0.4), but now I don't see any results under Experiments, and clicking update data on my archived experiment does not change the results (they are still from the variants yesterday. 2. When I turn on an experiment in growthbook it takes almost 2 hours for it to effect traffic on production. Is this expected? Have you guys seen this before? 3. If I want to test a new feature on 5% of traffic, is it better statistics to run (a) a test with 5% treatment, 95% control or (b) 10% of traffic, 50% treatment, 50% control.

Hello Everyone, we started to experiment with the GrowthBook. We would not want the rules to be exposed to the frontend ReactJS app. Wondering if there is any server side sdk that we can use to return only results of features to the frontend growthbook library

Does growthbook support split urls tests?