https://www.growthbook.io/ logo
Join Slack
Powered by
Hello there I looks like we overlooked something a...
# ask-questions
n
Hello there I looks like we overlooked something and it goes wrong. General question: should we change something in the code in order to use Namespaces function? Active test Groups A (33,3%), B (33,3%), C (33,3%). Namespace: [0, 0.5] Disabled test Groups A, B, Namespace [0.5, 1] In the Active test we have A: 80k users, B: 20k users, C: 20k users. (based on analytics) The split is not correct. Could you please advise where we can find the issue?
f
Hi Alexander - did you always have that same split or did you adjust it at some point?
n
Hi Graham - Sometimes we have two groups for this feature flag, sometimes we have three groups.
f
oh, for the same experiment name?
n
yes. But it was disabled before the latest run
f
the ‘track by’ is the same for both rules?
n
Do you mean this?
In the Active test we have one key (feature flag) and in the disabled test we have another key (feature flag)
The namespace has the same name in both tests
And we user deviceID for the splits on the device.
Based on numbers I can assume that Group A = Group A from the Namespace: [0, 0.5] + the whole Namespace [0.5, 1] - 80k Group B = Group B from Namespace: [0, 0.5] - 20k Group C = Group C from Namespace: [0, 0.5] - 20k
f
Did you add a new phase to the experiment to ignore past users?
(It will set a date range)
n
How can I add a new phase? We have very low retention rate, so the majority of the users are new.
And we don't want to ignore the past users. I thought GB splits every user every time it sees him.
I removed namespaces and I'm using 100% of the traffic. Now I have a nice 33% split for every group in the active test again
f
ok
n
But it doesn't solve the initial issue. 😭 We'd like to have mutually exclusive tests. I would greatly appreciate your advise, Graham.
f
Can you DM me a screen shot of the rules? Something seems a bit off
n
DM'd
38 Views
Open in Slack
PreviousNext
Powered by