Hi Andrea,
1. the activation metric is not aware of the splitting at all, so it is possible to get SRM errors from this - though we typically rarely see this as the assignment is random, the activation metric should be equally distributed.
2. The triggers, or targeting rules as we call them, are entirely custom. Whatever data is available at the time of assignment can be used.
3. We have this coming very soon with the Proxy, which will allow remote evaluation of flags. Currently to unlock this behaviour you would need to either pass this data down to the SDK level, or make a call to the data yourself. Though if you just want to use it as an activation metric, that is already supported as long as the event is in the same data warehouse as the exposure event.