Attempting to reason *carefully* about decision theory

I have been failing to notice my confusion about several aspects of decision theory. I am going to attempt to rectify this.

I am something that can perceive sensory data.

I notice that various sensory experiences are highly correlated.

I construct a large causal network from these correlations. This is a model of the world.

I notice that I have certain types of sensory experience that I can cause directly.

I notice that my sensory inputs are correlated with my caused outputs.

I add myself to the world model as a causal node.

I desire some sorts of sensory data over others.

I look for the sensory data I desire in my world model and work backwards, creating a plausible causal chain from the sensory input I desire to my current sensory input.

This screens off large portions of the space of all possible actions I have available to me.

Distinctions between "myself" and "the world" are a convenience for keeping track of things I can cause directly and things I can not.

This at least seems how the part of our mind we have access to reasons, what about the part we don't have access to? If it works forwards instead of backwards then presumably it screens off the action space and prunes the decision tree based on past correlations.

Hitting wall, must find sledgehammer, will email some people.