Control systems do not, in general, work by predicting what action will produce the intended perception. They need not make any predictions at all, nor contain any model of their environment. They require neither utility measures, nor Bayesian or any other form of inference. There are methods of designing control systems that use these concepts but they are not inherent to the nature of control.
But they should act according to a rigorous decision theory, even though they often don't. It seems to be an elementary enough statement, so I'm not sure what are you asserting.
"Should" statements cannot be logically derived from factual statements. Population evolution leads to evolutionarily stable strategies, not coherent decision theories.
I've often had half-finished LW post ideas and crossed them off for a number of reasons, mostly they were too rough or undeveloped and I didn't feel expert enough. Other people might worry their post would be judged harshly, or feel overwhelmed, or worried about topicality, or they just want some community input before adding it.
So: this is a special sort of open thread. Please post your unfinished ideas and sketches for LW posts here as comments, if you would like constructive critique, assistance and checking from people with more expertise, etc. Just pile them in without worrying too much. Ideas can be as short as a single sentence or as long as a finished post. Both subject and presentation are on topic in replies. Bad ideas should be mined for whatever good can be found in them. Good ideas should be poked with challenges to make them stronger. No being nasty!