You may want to check out https://en.wikipedia.org/wiki/Causal_inference https://en.wikipedia.org/wiki/Causal_inference - the issues that you point out are all

You may want to check out https://en.wikipedia.org/wiki/Causal_inference https://en.wikipedia.org/wiki/Causal_inference - the issues that you point out are all well understood and being addressed as well as possible within the limitations of our current knowledge of statistics, the logical fundamentals and the available data. Ignoring ordering for example isn't a problem with stochastic modelling per se, but just with this specific model when applied in a specific context - you could of course use a different model to account for ordering as well. Similar for whether you want to optimize the expected average outcome or some other statistic like a given percentile. Those models exist already, don't blame the discipline for your lack of knowledge.