Hello Team, We are using GrowthBook with the following setup: • Shopify storefront • GrowthBook Auto SDK (

auto.min.js

) •

data-tracking="growthbook"

• Managed Warehouse • Assignment identifier:

device_id

We observed unexpected behaviour regarding experiment assignments and would appreciate clarification. Scenario We created a brand new feature experiment with: • Production enabled • Experiment status = Running • Traffic allocation = 100% • No targeting rules • No code implementation in Shopify • No calls to

getFeatureValue()

,

evalFeature()

,

isOn()

, or any other SDK evaluation methods • No UI rendering based on the feature Despite this, the experiment immediately started generating rows in the

experiment_views

table and accumulating Device IDs in the Results page. For example: •

test

experiment generated 1,359 experiment views without any implementation. • Another non-implemented experiment (

slow_label_experiment

) generated 43,679 experiment views. We verified that these entries originate from the JS SDK (

sdk_language = js

,

sdk_version = 1.6.5

). Questions 1. Is it expected behavior for the Auto SDK with

data-tracking="growthbook"

to generate

experiment_views

or assignments for experiments that are never explicitly evaluated in application code? 2. Does the Auto SDK implicitly evaluate experiments present in the payload, even without calls to

getFeatureValue()

,

evalFeature()

, or

isOn()

? 3. Is this behavior specific to Managed Warehouse integrations? 4. What is the recommended workflow for teams that need to: ◦ create experiments before implementation, ◦ validate them in development and UAT environments, ◦ and avoid production Device IDs accumulating before launch? Our current workflow relies on creating experiments before implementation in order to complete QA and UAT validation, so keeping experiments in Draft or Traffic = 0 is not a viable option for us. We would appreciate guidance on the recommended setup for this use case. Thank you.

is it a feature flag or a visual experiment?

Feature Flag

how do you have an experiment on the feature flag without a rule? It should be a rule