I am running my first test in production and am starting to see Multiple Exposures Warning after some days. The code was off for some days due to some deployment. We are using the mpid (mParticles CDP ID for all users) as id in the Data Source. The test is also setting 3 individual test on the same page because multiple elements are changed. They do all have the same feature id and variant id´s How do I find the error?

Multiple exposures can be caused by a few things: • If the variation split changed during the experiment • If the assignment attribute and the analysis attribute (randomization unit) are different • if the id's used are not available before the assignment happens, and then the user might be quickly assigned to both as things load

The variation split is the same. There is no special targeting. The mpid can change in the session if e.g. the users logs in during the session. But if the ID is mpid for Growthbook then Growthbook would also see that as two different users, right? Or is there also som SDK id going on behind the scenes?

Hi Kasper, thanks very much for your patience on this thread. Is the mpid assigned or known only at the time the user logs in? If so, is it the case that a user is not uniquely identifiable until they log in and start a session (which logs their mpid)? I'm wondering if you're capturing data for users before they are uniquely identifiable. In this case, all those users would have an id of ​`null`​ before they log in, and all those ​`null`​ users could be causing a Multiple Exposures Warning, since GrowthBook would see many users with id ​`null`​ as viewing >1 experiment.

> But if the ID is mpid for Growthbook then Growthbook would also see that as two different users, right? Means that mParticle always gives an mpid (the id we use in Growthbook). But when the user then logs in the session could shift from the user being one mpid to being another mpid. I guess that means that the new logged in mpid user then is completely new in growthbook and havent necessarily seen the experiment yet. And when they see it they could easily be assigned another variant group. And because Growthbook doesnt have any alternative ID´s behind the scene this would not be seen as "Multiple Exposure". The Multiple Exposure did not start until after the Growthbook implementaiton was somehow of in the code - and we then fixed the code error. My hope is thus that we will not se this error on our next test.