training bayes and newsletters

Hi,

I've just retrained my bayes database (stored in SQL) with 10k hams and
about 6k spams. I tried to make sure there were no newsletters in either
corpus, but some emails present as newsletters but really are spam.
However, many legitimate newsletters are hitting BAYES_99 even though I
haven't trained them.

I can imagine the newsletter template is somewhat common, but does bayes
have any ability to distinguish a junk newsletter from a legitimate
newsletter? I realize there's somewhat of an imbalance between hams and
spams, but shouldn't there be enough?

Would I benefit from training known trustworthy newsletters such as ham?

Do you have any recommendations outside of adding the legitimate
newsletters hitting BAYES_99 to an allowlist?

My previous bayes database had 100k emails of each type, created over like
ten years, but also had problems with identifying newsletters properly.

Thanks,
Alex