Sure, but you can't simultaneously hold P(red) = 0.33 and P(red) = 0.3. Yet people asked to make two choices will strictly prefer "red" but also strictly prefer

Sure, but you can't simultaneously hold P(red) = 0.33 and P(red) = 0.3. Yet people asked to make two choices will strictly prefer "red" but also strictly prefer "black or yellow". If red is a better gamble, then red or black must be a better gamble too! It's the same urn. You're right in general that uncertainty aversion has been put forward as an explanation. But to model it, you have to break EUT.