Hi guys, One doubt about the phases in an experim...
# ask-questions
d
Hi guys, One doubt about the phases in an experiment. We are running an A/B test on Android, but we realised the experiment setup was incorrect and the balance was wrong. After fixing the configuration, we started a new phase with re-randomized data, and we set-up the correct date for checking the data result after the new phase. But we are still seeing the same balance issue in the experiment. PS: I have checked the code and the tracking, and it's working fine. Do you have any clue what could happen with the experiment and the balance?
f
its hard to debug this from here
but i would suggest adjusting the time frame for the beginning of the phase
and see if that helps
or look at the assignment over time graph to see if the balance is better recently
d
we did the adjustment of the time frame to sync with the new phase, but the issue is still there. We have stopped the experiment and forced the FT to the default value to clean the data and the device cache, and we will enable the experiment again next week with the correct configuration. Anyways, this experiment is conditional to the UI, so we are only sending the tracking experiment (control group, variant 1) when the user scrolls down to a specific part of the screen. Maybe this condition could affect the data? And thanks @fresh-football-47124 for your time 🙂
Hi @fresh-football-47124, We have restarted the experiment from the beginning, and after some time, we have the same issue. One question to confirm my hypothesis. In Android, we are using the GrowthBook SDK, and for this experiment, in the moment the user enter into the screen, we are calling the SDK function:
Copy code
fun feature(id: String): GBFeatureResult {
    val evaluator = GBFeatureEvaluator(
        createEvaluationContext(), this.forcedFeatures,
    )
    return evaluator.evaluateFeature(
        featureKey = id,
        attributeOverrides = attributeOverrides,
    )
}
To know in which variant the user is to know how to render in compose a part of the screen. But we are tracking the experiment in the moment the user scroll down in this screen and see the UI component. This behaviour could affect the SRM? Because can happen that we are consuming the experiment calling the feature function, but not sending the experiment tracking ("_assign_experiment_") if the user not scroll down.