does 0.9.1.2 need new training db?

[Barry Gould]

At 04:39 AM 12/8/2002, Matthias Andree wrote:
>With identical .db files? If so, we might have a first significant
>precedence of rounding error propagation...

I noticed last week (with 0.9.1.2CVS), that if you start from scratch, and 
insert a word like "fred" into BOTH databases, and then run bogofilter, it 
does NOT return 0.5:

>1003 > \rm *
>root at SalesAdmin /root/.bogofilter
>1004 > bogofilter -s
>fred^M^D
>root at SalesAdmin /root/.bogofilter
>1005 > bogofilter -n
>fred^M^D
>root at SalesAdmin /root/.bogofilter
>1006 > ll
>total 16
>-rw-r--r--    1 root     root         8192 Dec  9 10:04 goodlist.db
>-rw-r--r--    1 root     root         8192 Dec  9 10:04 spamlist.db
>root at SalesAdmin /root/.bogofilter
>1007 > bogofilter -vvv
>fred^M^D
>X-Bogosity: No, tests=bogofilter, spamicity=0.499958, 
>version=0.9.1.2.cvs.20021206
>      Token                    pgood      pbad        fw  invfwlog     fwlog
>   1  fred                      1.00         1  0.499958  -0.69306  -0.69323
>   2  P_Q_S_invsum_logsum    0.49996   0.50004  0.499958    -0.693    -0.693