0.17.2 and 0.17.3 differences in scoring

[Tom Allison]

block_on_subnets=no

I came in this AM and found 5 Unsure emails which was atypical.

So I decided to run these 5 through bogofilter -vv and see what came up.
Then I reverted back to v 0.17.2 and did it again.
0.17.2 caught two of the 5.  the real answer was 5 ham, 0 spam.

I posted on pair of -vv output at the bottom.  I didn't go with -vvv 
because I didn't want to get too verbose on the mailing list.

For my own sanity, can someone expand on the lexer change?  I didn't see 
it in the Release Notes.  Also, what's a good way to activate this 
variable for transactional processing?

Currently I run all my Input to bogofilter's wordlist through shell 
scripts (train on error, train to exhaustion) and I supposed I could 
just declare it in there.  Yes?



1079695338.19457.janus:2,
X-Bogosity: Unsure, tests=bogofilter, spamicity=0.181186, version=0.17.3
    int  cnt   prob  spamicity histogram
   0.00   22 0.056908 0.021090 ######################
   0.10   20 0.145840 0.053745 ####################
   0.20   14 0.245889 0.088345 ##############
   0.30   16 0.343408 0.138043 ################
   0.40    0 0.000000 0.138043
   0.50    0 0.000000 0.138043
   0.60    7 0.659962 0.187633 #######
   0.70   28 0.742852 0.342909 ############################
   0.80    7 0.858408 0.377630 #######
   0.90    9 0.957536 0.451672 #########

1079695338.19457.janus:2,
X-Bogosity: No, tests=bogofilter, spamicity=0.096032, version=0.17.2
    int  cnt   prob  spamicity histogram
   0.00   22 0.056908 0.021140 ######################
   0.10   20 0.145840 0.053910 ####################
   0.20   14 0.245889 0.088631 ##############
   0.30   16 0.343408 0.138482 ################
   0.40    0 0.000000 0.138482
   0.50    0 0.000000 0.138482
   0.60    7 0.659962 0.188167 #######
   0.70   28 0.740549 0.342725 ############################
   0.80    7 0.858408 0.377526 #######
   0.90    8 0.952471 0.443552 ########