Hey Team I've a question about reporting when star...
# ask-questions
f
Hey Team I've a question about reporting when starting a new phase We are evaluating GB to be used as A/B testing framework in our backend service. In our implementation, it could take 15-20 mins for the updates to reflect on all VMs of the service. But start time of the new phase would be immediate in GB, how do we avoid metrics for the previous phase to spill into the new phase. Eg. Phase 1 (A -> 50%, B-> 50%) March 24th 00:00 to March 25 01:00 Phase 2 (A -> 60%, B -> 40%) March 25th 01:00 to - Now let's say the new configs reflect in the backend service around March 25 01:15. GB would pull data for phase 2 from March 25 01:00, but first 15 mins were running phase 1. we get
experiment.getPhase()
in GB client SDK and that can be used to separate data for various phases in the reporting tables, is there any way to filter out data for a phase in GB Results tab using this field, or is date range the only option.
f
I would suggest that you set the phase to the latest possible time the changes will propagate after the change
this will lose some transitional data, but will ensure that the exposures are correct, and you'll avoid possible SRM errors
f
ok so this is what i have to do 1. On the experiment page -> Make changes -> start a new phase 2. Top right corner click on 3 dots -> edit phases -> change start time of the new phase to latest possible time.
f
yes - there are two sides to the experiment - assignment and reporting
on the experiment reporting side, you can set that date to whatever you like, it will just exclude traffic exposed before that time
🙌 1
f
Great, Thanks @fresh-football-47124