Maybe I am misunderstanding your issue, but I have the feeling that you consider that if LT does not detect any mistakes in a machine-translated sentence, then it is a good translation. Well, one of the greatest critic of Google Translate (for instance), is that it often generates correct sentences, but with a meaning totally different from the inital sentence. For instance, from « La muson manĝas la kato. » (meaning “The cat eats the mouse.” in Esperanto), I get “The cat eats the cat.” as a translation (source: https://translate.google.com/?source=osdd#auto/en/la%20muson%20man%C4%9Das%20la%20kato ). The machine-translated sentence is correct (although strange), but does not mean the same than the initial sentence.
My conclusion is that the A and B options are dangerous: however good the results of LT are, it can not be fully trusted, and a speaker of the language should check the sentence. It is thus a good hypothesis (I think) to consider that the users of LT speak the checked language. Option C seems an interesting feature to me.