> I'd say that manual training of spam and ham in such group is needed > in such cases.
manuel training is sign of to much whitelistning with default score of -100, most ham here does not even need it, so bayes is here is more passive on this here, and this makes more sure that ham that is ham is autolearned as ham, still have not needed to help bayes here to make the right scores, but i am avare of this as scores is not always best with -100, eg use def_whitelist_* if whitelist_from_* is to much when maillist groubs get in grey of ham spam then manuel training can be one way of make it, but olso its possible to make more rules to catch spam, its just a matter of prefer :)
