It seems that the domain sayzsw.com seems to be clean in regards to bayes and thus it levels out the score as can be observed below. Is there a way that I can tweak bayes to reset or set a customized score for an individual entry or to remove the entry entirely?
Not easily.. if you can dig in the archives, Theo has made several posts explaining how to unpack the bayes DB.
Basically I want to remove sayzsw.com from bayes so that this message would have been considered spam and not relearned as ham in bayes.
Also, one piece of small advice is you should realize that the only reason it would prevent it from learning would not be due to score, but due to the restriction of autolearn not contradicting existing learning. Autolearning is based on the message score without bayes included, so a positive message score resulting from bayes won't keep it from autolearning as ham unless it's a strong enough bayes score to trigger the "no contradictions" behavior.
Personally, my suggestion to fix this kind of thing in the longer-term is to set your autolearn threshold to -0.01, and add some custom rules that add tiny negative scores to emails containing phrases typical of nonspam for your site. This will prevent the "learn as ham anything that doesn't match any rules" behavior of SA's default autolearn thresholds (which IMHO are completely broken).
The only reason for the non-negative autolearn threshold is that all negative scoring comp rules, except habeas, were removed due to spammer abuse.. I'm of the opinion that very small negative scoring rules are valuable, because spammers can't abuse them for any significant benefit, and they help distinguish between non-spam and clever spam that manages to dodge all the positive scoring rules. just don't see much sense in the default of "learn as ham if it doesn't contain any known spam markers". My philiosophy is that learning as ham needs to be earned, not just a default.
