Bounce options for bayes

[Pete russell]

We are a Lotus Notes shop, once our users receive an email, that is
spam, that needs to be forwarded to a spam account, so we can run the
sa-learn bash script against it each night, the mail will have had it
headers re arranged by Lotus notes - I can see no way around this at the
notes end, and would like to create a local postfix account to receive
all this mail from my users, and then run Julian's sa-learn bash script
against it - but have some questions about setting this up.

1. Is there an easy way, if I create a new spam.assassin.prefs.conf, to
specify not to use any of the header info in bayes learning? I want to
create a separate set of bayes rules for this batch of mail, due to the
big change in header info, leaving the autolearning bayes settings
alone. I know how to change this in the bash script, but not what
contents to have in the prefs file, or if it is allowed to have more
than one?

2. I will use a person doc (much like a .forward file) to forward mail
sent by users to spam/notes (spam at mydomain.com) to a local postfix mail
account, that we can then run the sa-learn bash script against. Any
postfix gurus here know how I can easily specify in postfix to allow
only this one specific account to receive local mail, but no others? I
currently have local delivery off in the outbound postfix MTA, so would
need the inbound postfix MTA to deliver this locally without sending to
MailScanner.

Thanks for any suggestions - and for putting up with my noob questions

Pete