After a quick skim of https://plato.stanford.edu/entries/foucault/#HistPris I think I can relate the idea of power-knowledge to something with which I am much more familiar: closed loop control theory.
Under that model, it makes sense that one's ability to control would be dependent not only upon how good one's effectors may be ("power"?) but also upon how good one's sensors may be ("knowledge"?). (let alone questions of system identification!)
But that could easily be me using the wrong prism ("observability/controllability") for analysis.