I think that may be a separate experiment, as I want to focus solely on methods and agents that are focused on clarity.
I think I'll stick to one of the above methods for all batches, include cold crashing on one of them, and then once the clarity is accounted for, re-do the experiment with...