Is there any possibility that the expert action might have been compromised?

[paradoxlove9]

Given that the dataset combines suboptimal and expert data, including expert states and actions, and that the actions from the mixed dataset are used when optimizing the strategy, does this imply that the expert actions are utilized during the policy update?