The whole seeming thrust of this is to try to create a way to measure how good an answer is, and it seems based on the assumption that the best answer, or some other factors, will determine that an answer is correct.
You didn't ask what you mean by correct. If correct means the person who asked found it answered the question to their satisfaction, then that's at least in line with what SA is about, if not covering every case (or even the majority necessarily).
However, given that it seems that you are trying to train an AI aimed at "helping us understand traditional sources", we are left to guess that when you say "correct", you mean the answer is "the correct answer, as faithfully and authentically provided by the tradition". This is a big no no for SA.
SA doesn't have a haskama, and it is questions and answers from anons. There is no Rav on board showing every answer to their Rosh Kollel to verify it. You'll also be aware that the tradition is rich and varied, and there are often many answers that fit your criteria of "correct" on the same question.
So, I am not an AI programmer, I don't know if you can measure "trustworthiness" in how you rate the data you feed your bot, but I wouldn't know how to measure that for SA, but do note that it will certainly not be very high. Given what you are asking, I think it is impossible, and certainly not a good idea. If this is just an AI to look things up in a chat-like way, then fine (and if so, what makes it different to bing?), but if it is actually trying to understand this data as some sort of "Rav.AI" then my answer applies.
If you clarify the question in such a way that I feel you are asking something that can be answered, and takes into account the facts about how this site works, and how they affect your goals, I will happily retract my downvote and perhaps give an upvote.