When I assign credibility ratings ... I am careful to do it regardless of the correctness of the post...
No, you don't. Your "credibility ratings" correlate to whether the data are favorable to your belief. They correlate negatively to posts where the credibility is not even a matter of subjective judgment anymore, because you have been told by their authors whether they were intended seriously. You disregard this information and try to accuse the authors of revisionism -- but only in the cases where the data favor your belief.
I believe, and continue to believe that this is not only very doable, but also easily verifiable by readers who follow the work.
No. It's blatantly cheating, and the readers who have followed your work have
not agreed in the least with your attempts at
post hoc data filtration. Quite the contrary, they have recommended methods that make this
post hoc review entirely unnecessary, and would completely obviate any need to justify it or convince anyone that it's not improper manipulation of the data. When you refuse to consider such protocols, the readers rightly conclude that it's probably your intention to structure your experiment in a way that gives you opportunities to cheat, and then hope later not to get caught.
In a telepathy experiment, if friendly and serious-sounding answers...
You've proven you can't tell serious answers from facetious ones.
...tend to be more accurate than those which seem hostile, aggressive, absurd or crazy, it would be a monumentally foolish error to not separate out the friendly and serious-sounding answers...
No. If you're admitting that your opinion of the tone of answers is just a proxy for the "accuracy" of the answer, then you're just putting a thin smokescreen in front of value-based data culling, hoping to make it look like blind metadata-based culling. Believe me, we can see right through it.
I generally find that this procedure leads to a significant improvement...
Probably because that's what it was intended to do. And exactly why it's cheating.