You look at how much of the ultimate value get from achieving your alternative goals and choose the ones which in the long run gives you most of the value. Let’s suppose the ultimate value is happiness. We can then identify bottlenecks (like you suggest) to producing the most amount of happiness possible. A binary sub-goal you can identify as most important at one moment might be “get enough money to buy a home and feel secure”, at another point it could be “get a consistent sleep schedule”.
The sub-goals are not ends in themselves. So the point isn’t really to get the thing thing they’re about, but rather to get the ultimate value you eventually get using the outcome of this sub-goal. For example, getting the house isn’t the end in itself. We get the house such that we have a safe and comfortable living space which gives us happiness and also let’s us achieve more goals which give us happiness. Assuming happiness is the ultimate goal. So what I want to get at here is that the sub-goals wouldn’t be mixing qualitatively different values, but they would be measured by the amount of the same ultimate value they produce.
Logically, there could be multiple ultimate goals. That would be there were multiple things which are ends in themselves. In that case there’s no way to maximize all of them. At least I don’t see why that would be logical impossibility.
Elliot pointed out that the goal might not have to be maximized:
The thing about new and interesting ways could apply if the goal isn’t a quantity as well.