My local.cf is the same as yours.
I have amended the scores for certain test in my user_prefs file and SA is doing very well at the moment. Since last night it's blocked approximately 193 spam messages and let through 6.
My user_prefs file is as follows...
Code:
# SpamAssassin user preferences file
# Encapsulate spam in an attachment (0=no, 1=yes, 2=safe)
report_safe 0
# Enable the Bayes system
use_bayes 1
# Enable Bayes auto-learning
bayes_auto_learn 1
bayes_auto_learn_threshold_spam 9.0
bayes_auto_learn_threshold_nonspam -0.1
# Acceptable languages for emails
ok_languages en
ok_locales en
# Check Various DNSBLs
score RCVD_IN_MAPS_RBL 3.4
score RCVD_IN_MAPS_DUL 3.0
score RCVD_IN_MAPS_RSS 3.2
score RCVD_IN_MAPS_NML 2.0
score RCVD_IN_NJABL_DUL 3.0
score RCVD_IN_SORBS_DUL 3.0
score RCVD_IN_SORBS_WEB 1.0
score RCVD_IN_SBL 12.0
score RCVD_IN_XBL 12.0
score DCC_CHECK 2.0
score PYZOR_CHECK 3.0
score RAZOR2_CHECK 0
score RAZOR2_CF_RANGE_51_100 0
# Check message body for URLs
score URIBL_SBL 12.0
score URIBL_SC_SURB 3.0
score URIBL_WS_SURBL 1.0
score URIBL_PH_SURBL 2.0
score URIBL_OB_SURBL 3.0
score URIBL_AB_SURBL 2.0
score URIBL_JP_SURBL 2.5
# Amend weighting of porn emails
score SUBJECT_SEXUAL 12.0
score FREE_PORN 3.2
score CUM_SHOT 3.2
score LIVE_PORN 3.2
score HARDCORE_PORN 3.2
score HOT_NASTY 2.0
score BEST_PORN 3.2
score NASTY_GIRLS 3.2
score AMATEUR_PORN 3.2
score PORN_CELEBRITY 3.2
score PORN_15 3.2
score PORN_16 3.2
score PORN_URL_SEX 1.0
score PORN_URL_SLUT 1.0
# Amend weighting of emails with a very low text-image ratio
score HTML_IMAGE_ONLY_04 3.0
score HTML_IMAGE_ONLY_08 3.0
# Amend SPF weighting (doesn't appear to work at UH at the moment)
score SPF_PASS -1.0
score SPF_FAIL 3.0
score SPF_SOFTFAIL 1.0
score SPF_HELO_PASS -1.1
score SPF_HELO_FAIL 3.0
score SPF_HELO_SOFTFAIL 1.0
# Amend Bayesian Scores
score BAYES_99 10
score BAYES_95 3.4
score BAYES_80 3.0
score BAYES_60 2.5
score BAYES_50 2.2
score BAYES_40 1.0
score BAYES_20 0.3
score BAYES_05 -0.5
score BAYES_00 -2.0
# Set required score for emails to be classed as spam
required_score 3.5
# Whitelist domains
whitelist_from *@somedomain.co.uk
# Leave email subject alone
rewrite_header Subject
I've whitelisted as many accounts or domains as I can which has reduced the false-positives to nothing thus far. I also don't rewrite the email subject to include [Spam] but move it to a folder called Junk.
Furthermore, I run sa-learn once a day or so on both my Inbox and Junk folders (so SA learns both what is and isn't spam). The Bayesian scoring is now pretty accurate. The number of false-negatives that are reaching my Inbox has been creeping up the past couple of days so I may need to tweak some more.
Ben
Linear Mode
