How does separating chrome change the maths here? Those scripts are incredibly pervasive so would obviously be candidates for this making a performance impact. Using Google URLs for the experiment seems fair enough given they can verify they meet the conditions explained in the design doc?