that old message I was talking about.

---------- Forwarded message ----------
From: Daniel Quinlan <quin...@pathname.com>
Date: Sat, May 22, 2004 at 16:25
Subject: DNSBL accuracy using -firsttrusted
To: spamassassin-...@incubator.apache.org


Someone at Spamhaus poked me to try testing only the last IP address
with XBL and I tested it and it helps reduce false positives quite
nicely.  The concept with XBL is that if it came most recently from an
okay host, then the message is probably okay too.  It's a bit spooky but
it works and I suppose it is closer in behavior to how blacklists are
generally used at connect time, so perhaps most are tuned to be used
this way.

The main caveat is that if trusted networks is not guessed or set
correctly, then *no* blacklist hits will happen and the net score set
will be used to the detriment of the site.

I tried the same idea on more or less every applicable blacklist and
check out the results:

------- start of cut text --------------
OVERALL%   SPAM%     HAM%     S/O    RANK   SCORE  NAME
 29979    14999    14980    0.500   0.00    0.00  (all messages)
100.000  50.0317  49.9683    0.500   0.00    0.00  (all messages as %)

 12.212  24.4083   0.0000    1.000   1.00    0.01  T_RCVD_IN_NJABL_PROXY
 12.962  25.7951   0.1135    0.996   0.57    0.00  RCVD_IN_NJABL_PROXY

 18.186  36.3291   0.0200    0.999   0.95    1.00  __T_RCVD_IN_NJABL
 19.877  38.1225   1.6088    0.960   0.30    1.00  __RCVD_IN_NJABL

 8.613  17.2145   0.0000    1.000   0.91    0.01  T_RCVD_IN_SORBS_MISC
 9.136  18.2412   0.0200    0.999   0.80    0.00  RCVD_IN_SORBS_MISC

 29.124  58.1705   0.0401    0.999   0.90    0.01  T_RCVD_IN_DSBL
 30.395  60.2640   0.4873    0.992   0.43    0.00  RCVD_IN_DSBL

 7.966  15.9211   0.0000    1.000   0.87    0.01  T_RCVD_IN_SORBS_HTTP
 8.449  16.8011   0.0868    0.995   0.49    0.00  RCVD_IN_SORBS_HTTP

 5.337  10.6540   0.0134    0.999   0.74    0.01  T_RCVD_IN_RFCI
 7.162  12.3675   1.9493    0.864   0.00    0.00  RCVD_IN_RFCI

 9.804  19.5613   0.0334    0.998   0.73    0.01  T_RCVD_IN_SBL
 9.927  19.7747   0.0668    0.997   0.62    0.00  RCVD_IN_SBL

 14.610  29.1486   0.0534    0.998   0.73    1.00  __T_RCVD_IN_SBL_XBL
 15.044  29.7820   0.2870    0.990   0.35    1.00  __RCVD_IN_SBL_XBL

 3.116   6.2204   0.0067    0.999   0.72    0.00  RCVD_IN_NJABL_SPAM
 3.062   6.1137   0.0067    0.999   0.70    0.01  T_RCVD_IN_NJABL_SPAM

 2.055   4.1069   0.0000    1.000   0.66    0.01  T_RCVD_IN_BL_SPAMCOP_NET
 2.235   4.3070   0.1602    0.964   0.14    0.00  RCVD_IN_BL_SPAMCOP_NET

 5.100  10.1740   0.0200    0.998   0.65    0.01  T_RCVD_IN_XBL
 5.417  10.6074   0.2203    0.980   0.18    0.00  RCVD_IN_XBL

 21.869  43.5562   0.1535    0.996   0.64    0.01  T_RCVD_IN_SORBS_DUL
 22.146  44.0363   0.2270    0.995   0.48    0.00  RCVD_IN_SORBS_DUL

 34.071  67.9112   0.1869    0.997   0.63    1.00  __T_RCVD_IN_SORBS
 42.410  70.9047  13.8785    0.836   0.34    1.00  __RCVD_IN_SORBS

 1.868   3.7336   0.0000    1.000   0.64    0.00  RCVD_IN_SORBS_SMTP
 1.731   3.4602   0.0000    1.000   0.62    0.01  T_RCVD_IN_SORBS_SMTP

 2.935   5.8537   0.0134    0.998   0.63    0.00  RCVD_IN_NJABL_DIALUP
 2.879   5.7404   0.0134    0.998   0.61    0.01  T_RCVD_IN_NJABL_DIALUP

 0.934   1.8668   0.0000    1.000   0.57    0.01  T_RCVD_IN_RSL
 1.041   2.0735   0.0067    0.997   0.55    0.00  RCVD_IN_RSL

 0.607   1.2134   0.0000    1.000   0.53    0.01  T_RCVD_IN_SORBS_SOCKS
 0.637   1.2401   0.0334    0.974   0.33    0.00  RCVD_IN_SORBS_SOCKS

 0.430   0.8601   0.0000    1.000   0.49    0.01  T_RCVD_IN_SORBS_WEB
 0.447   0.8867   0.0067    0.993   0.46    0.00  RCVD_IN_SORBS_WEB

 0.254   0.5067   0.0000    1.000   0.44    0.01  T_RCVD_IN_SORBS_ZOMBIE
 0.307   0.5867   0.0267    0.956   0.29    0.00  RCVD_IN_SORBS_ZOMBIE

 0.117   0.2333   0.0000    1.000   0.42    0.00  RCVD_IN_NJABL_RELAY
 0.113   0.2267   0.0000    1.000   0.40    0.01  T_RCVD_IN_NJABL_RELAY
------- end ----------------------------

change in RANK (relative to just the IP-based blacklists and the new
-firsttrusted ones in testing)

  0.74   RCVD_IN_RFCI
  0.52   RCVD_IN_BL_SPAMCOP_NET
  0.47   RCVD_IN_XBL
  0.47   RCVD_IN_DSBL
  0.43   RCVD_IN_NJABL_PROXY
  0.38   RCVD_IN_SORBS_HTTP
  0.20   RCVD_IN_SORBS_SOCKS
  0.16   RCVD_IN_SORBS_DUL
  0.15   RCVD_IN_SORBS_ZOMBIE
  0.11   RCVD_IN_SORBS_MISC
  0.11   RCVD_IN_SBL
  0.03   RCVD_IN_SORBS_WEB
  0.02   RCVD_IN_RSL
 -0.02   RCVD_IN_NJABL_DIALUP
 -0.02   RCVD_IN_NJABL_RELAY
 -0.02   RCVD_IN_NJABL_SPAM
 -0.02   RCVD_IN_SORBS_SMTP

and not really relevant unless we change entire sets to reduce the
number of look-ups:

  0.65   __RCVD_IN_NJABL
  0.38   __RCVD_IN_SBL_XBL
  0.29   __RCVD_IN_SORBS

Results for some fresh mail that may still have a few misfiles:

------- start of cut text --------------
OVERALL%   SPAM%     HAM%     S/O    RANK   SCORE  NAME
  4039     2294     1745    0.568   0.00    0.00  (all messages)
100.000  56.7962  43.2038    0.568   0.00    0.00  (all messages as %)

 48.651  85.6582   0.0000    1.000   1.00    1.00  __T_RCVD_IN_SBL_XBL
 50.012  87.8378   0.2865    0.997   0.68    1.00  __RCVD_IN_SBL_XBL

 34.043  59.9390   0.0000    1.000   0.93    0.01  T_RCVD_IN_BL_SPAMCOP_NET
 35.058  61.4211   0.4011    0.994   0.55    0.00  RCVD_IN_BL_SPAMCOP_NET

 33.028  58.1517   0.0000    1.000   0.90    0.01  T_RCVD_IN_XBL
 34.315  60.3313   0.1146    0.998   0.78    0.00  RCVD_IN_XBL

 28.571  50.3051   0.0000    1.000   0.86    1.00  __T_RCVD_IN_SORBS
 36.172  53.0078  14.0401    0.791   0.11    1.00  __RCVD_IN_SORBS

 26.195  46.1203   0.0000    1.000   0.81    0.01  T_RCVD_IN_DSBL
 27.309  47.2537   1.0888    0.977   0.32    0.00  RCVD_IN_DSBL

 19.064  33.5658   0.0000    1.000   0.78    0.00  RCVD_IN_SORBS_DUL
 18.693  32.9119   0.0000    1.000   0.76    0.01  T_RCVD_IN_SORBS_DUL

 16.712  29.4246   0.0000    1.000   0.71    0.01  T_RCVD_IN_SBL
 16.984  29.7733   0.1719    0.994   0.48    0.00  RCVD_IN_SBL

 11.290  19.8779   0.0000    1.000   0.67    1.00  __T_RCVD_IN_NJABL
 12.528  20.6190   1.8911    0.916   0.00    1.00  __RCVD_IN_NJABL

 7.130  12.5545   0.0000    1.000   0.62    0.01  T_RCVD_IN_NJABL_PROXY
 7.576  13.2084   0.1719    0.987   0.38    0.00  RCVD_IN_NJABL_PROXY

 5.967  10.5057   0.0000    1.000   0.57    0.01  T_RCVD_IN_RFCI
 7.774  12.5545   1.4900    0.894   0.01    0.00  RCVD_IN_RFCI

 5.769  10.1569   0.0000    1.000   0.55    0.01  T_RCVD_IN_SORBS_MISC
 6.041  10.5929   0.0573    0.995   0.51    0.00  RCVD_IN_SORBS_MISC

 4.481   7.8901   0.0000    1.000   0.50    0.01  T_RCVD_IN_SORBS_HTTP
 4.828   8.3697   0.1719    0.980   0.26    0.00  RCVD_IN_SORBS_HTTP

 2.253   3.9669   0.0000    1.000   0.47    0.01  T_RCVD_IN_NJABL_DIALUP
 2.253   3.9669   0.0000    1.000   0.47    0.00  RCVD_IN_NJABL_DIALUP

 2.179   3.8361   0.0000    1.000   0.45    0.00  RCVD_IN_RSL
 1.956   3.4438   0.0000    1.000   0.43    0.01  T_RCVD_IN_RSL

 1.783   3.1386   0.0000    1.000   0.40    0.01  T_RCVD_IN_NJABL_SPAM
 1.783   3.1386   0.0000    1.000   0.40    0.00  RCVD_IN_NJABL_SPAM

 0.693   1.2206   0.0000    1.000   0.38    0.00  RCVD_IN_SORBS_SOCKS
 0.693   1.2206   0.0000    1.000   0.38    0.01  T_RCVD_IN_SORBS_SOCKS

 0.619   1.0898   0.0000    1.000   0.35    0.00  RCVD_IN_SORBS_SMTP
 0.569   1.0026   0.0000    1.000   0.33    0.01  T_RCVD_IN_SORBS_SMTP

 0.545   0.9590   0.0000    1.000   0.31    0.00  RCVD_IN_SORBS_WEB
 0.545   0.9590   0.0000    1.000   0.31    0.01  T_RCVD_IN_SORBS_WEB

 0.149   0.2616   0.0000    1.000   0.26    0.01  T_RCVD_IN_NJABL_RELAY
 0.149   0.2616   0.0000    1.000   0.26    0.00  RCVD_IN_NJABL_RELAY

 0.099   0.1744   0.0000    1.000   0.23    0.00  RCVD_IN_SORBS_ZOMBIE
 0.074   0.1308   0.0000    1.000   0.21    0.01  T_RCVD_IN_SORBS_ZOMBIE
------- end ----------------------------

  0.56   RCVD_IN_RFCI
  0.49   RCVD_IN_DSBL
  0.38   RCVD_IN_BL_SPAMCOP_NET
  0.24   RCVD_IN_SORBS_HTTP
  0.24   RCVD_IN_NJABL_PROXY
  0.23   RCVD_IN_SBL
  0.12   RCVD_IN_XBL
  0.04   RCVD_IN_SORBS_MISC
  0.00   RCVD_IN_SORBS_WEB
  0.00   RCVD_IN_SORBS_SOCKS
  0.00   RCVD_IN_NJABL_SPAM
  0.00   RCVD_IN_NJABL_RELAY
  0.00   RCVD_IN_NJABL_DIALUP
 -0.02   RCVD_IN_RSL
 -0.02   RCVD_IN_SORBS_DUL
 -0.02   RCVD_IN_SORBS_SMTP
 -0.02   RCVD_IN_SORBS_ZOMBIE

  0.75   __RCVD_IN_SORBS
  0.67   __RCVD_IN_NJABL
  0.32   __RCVD_IN_SBL_XBL

and now Justin's results:

------- start of cut text --------------
OVERALL%   SPAM%     HAM%     S/O    RANK   SCORE  NAME
 15995     7997     7998    0.500   0.00    0.00  (all messages)
100.000  49.9969  50.0031    0.500   0.00    0.00  (all messages as %)

 11.279  22.5585   0.0000    1.000   0.94    0.01  T_RCVD_IN_NJABL_PROXY
 12.035  23.5338   0.5376    0.978   0.77    1.20  RCVD_IN_NJABL_PROXY

 16.211  32.4247   0.0000    1.000   0.96    0.00  __T_RCVD_IN_NJABL
 19.425  33.8502   5.0013    0.871   0.60    0.00  __RCVD_IN_NJABL

 8.234  16.4687   0.0000    1.000   0.91    0.01  T_RCVD_IN_SORBS_MISC
 9.015  17.2940   0.7377    0.959   0.71    1.20  RCVD_IN_SORBS_MISC

 32.873  65.7497   0.0000    1.000   0.99    0.01  T_RCVD_IN_DSBL
 34.117  67.5628   0.6752    0.990   0.79    1.10  RCVD_IN_DSBL

 7.984  15.9685   0.0000    1.000   0.90    0.01  T_RCVD_IN_SORBS_HTTP
 8.434  16.7063   0.1625    0.990   0.85    1.20  RCVD_IN_SORBS_HTTP

 7.265  14.5304   0.0000    1.000   0.90    0.01  T_RCVD_IN_RFCI
 8.778  16.5062   1.0503    0.940   0.66    0.10  RCVD_IN_RFCI

 2.738   5.4771   0.0000    1.000   0.80    0.01  T_RCVD_IN_SBL
 3.026   5.9772   0.0750    0.988   0.78    1.27  RCVD_IN_SBL

 39.162  78.3294   0.0000    1.000   1.00    0.00  __T_RCVD_IN_SBL_XBL
 40.913  80.7428   1.0878    0.987   0.75    0.00  __RCVD_IN_SBL_XBL

 0.863   1.7256   0.0000    1.000   0.63    0.01  T_RCVD_IN_NJABL_SPAM
 0.913   1.7757   0.0500    0.973   0.62    0.74  RCVD_IN_NJABL_SPAM

 11.597  23.1962   0.0000    1.000   0.94    0.01  T_RCVD_IN_BL_SPAMCOP_NET
 12.085  23.8089   0.3626    0.985   0.83    2.25  RCVD_IN_BL_SPAMCOP_NET

 37.149  74.3029   0.0000    1.000   0.99    0.01  T_RCVD_IN_XBL
 38.700  76.3911   1.0128    0.987   0.76    1.00  RCVD_IN_XBL

 23.370  46.7425   0.0000    1.000   0.97    0.01  T_RCVD_IN_SORBS_DUL
 23.651  46.9926   0.3126    0.993   0.87    2.55  RCVD_IN_SORBS_DUL

 33.117  66.2373   0.0000    1.000   0.99    0.00  __T_RCVD_IN_SORBS
 49.722  69.5011  29.9450    0.699   0.54    0.00  __RCVD_IN_SORBS

 2.588   5.1519   0.0250    0.995   0.78    1.20  RCVD_IN_SORBS_SMTP
 2.501   5.0019   0.0000    1.000   0.78    0.01  T_RCVD_IN_SORBS_SMTP

 3.532   7.0651   0.0000    1.000   0.83    0.62  RCVD_IN_NJABL_DIALUP
 3.482   6.9651   0.0000    1.000   0.83    0.01  T_RCVD_IN_NJABL_DIALUP

 4.020   8.0405   0.0000    1.000   0.85    0.53  RCVD_IN_RSL
 3.732   7.4653   0.0000    1.000   0.84    0.01  T_RCVD_IN_RSL

 0.706   1.4130   0.0000    1.000   0.61    0.01  T_RCVD_IN_SORBS_SOCKS
 0.706   1.4130   0.0000    1.000   0.61    1.62  RCVD_IN_SORBS_SOCKS

 0.688   1.3755   0.0000    1.000   0.61    0.01  T_RCVD_IN_SORBS_WEB
 0.719   1.4255   0.0125    0.991   0.60    2.90  RCVD_IN_SORBS_WEB

 0.088   0.1751   0.0000    1.000   0.50    2.70  RCVD_IN_SORBS_ZOMBIE
 0.025   0.0500   0.0000    1.000   0.49    0.01  T_RCVD_IN_SORBS_ZOMBIE

 0.688   1.3755   0.0000    1.000   0.61    0.01  T_RCVD_IN_NJABL_RELAY
 0.719   1.4005   0.0375    0.974   0.59    1.41  RCVD_IN_NJABL_RELAY

 0.000   0.0000   0.0000    0.500   0.49    0.10  RCVD_IN_NJABL_MULTI
 0.000   0.0000   0.0000    0.500   0.49    0.01  T_RCVD_IN_NJABL_MULTI

 0.000   0.0000   0.0000    0.500   0.49    0.01  T_RCVD_IN_SORBS_BLOCK
 0.000   0.0000   0.0000    0.500   0.49    0.00  RCVD_IN_SORBS_BLOCK

 0.000   0.0000   0.0000    0.500   0.49    0.01  T_RCVD_IN_NJABL_CGI
 0.000   0.0000   0.0000    0.500   0.49    0.10  RCVD_IN_NJABL_CGI
------- end ----------------------------

This RANK relative to all tests since this is from last night's run:

  0.24   RCVD_IN_RFCI
  0.23   RCVD_IN_XBL
  0.20   RCVD_IN_SORBS_MISC
  0.20   RCVD_IN_DSBL
  0.17   RCVD_IN_NJABL_PROXY
  0.11   RCVD_IN_BL_SPAMCOP_NET
  0.10   RCVD_IN_SORBS_DUL
  0.05   RCVD_IN_SORBS_HTTP
  0.02   RCVD_IN_SBL
  0.02   RCVD_IN_NJABL_RELAY
  0.01   RCVD_IN_SORBS_WEB
  0.01   RCVD_IN_NJABL_SPAM
  0.00   RCVD_IN_SORBS_SOCKS
  0.00   RCVD_IN_SORBS_SMTP
  0.00   RCVD_IN_SORBS_BLOCK
  0.00   RCVD_IN_NJABL_MULTI
  0.00   RCVD_IN_NJABL_DIALUP
  0.00   RCVD_IN_NJABL_CGI
 -0.01   RCVD_IN_RSL
 -0.01   RCVD_IN_SORBS_ZOMBIE

  0.45   __RCVD_IN_SORBS
  0.36   __RCVD_IN_NJABL
  0.25   __RCVD_IN_SBL_XBL

It seems like a huge improvement across the board.  There are a few
small slips, but do we care?

--
Daniel Quinlan
http://www.pathname.com/~quinlan/

Reply via email to