[I am not sure about this being covered by the TODO
10) New Feature: Add support for a user configurable
list of headers that should be used to ignore (single
or multi-line) headers that appear in the list. The
list should be used to ignore headers both during the
message registration and evaluation procedures.
But I believe its not, because I am concerned just in
the work done with the options -S/-N and no other]
* * *
I am testing bogofilter version 0.7.5.1 (from Debian sid)
I filter my email through bogofilter -u -e -p
This adds a header like
X-Bogosity: No, tests=bogofilter, spamicity=0.000043,
to my email. That is all right.
When the email classification by bogofilter is wrong
(false positives or negatives), I feed back the message
to bogofilter with bogofilter -N or bogofilter -S.
Everything seems preety nice, except that the message
sent back to bogofilter to correct its statistics is
different from the original one, since it has the new
Because of that, the whole statistical data becomes
slightly poluted by the words present in that header.
Thats not a terrible thing, but...
I dont know what would be the better strategy to
A possible one would be to ignore the header X-Bogosity
when processing a message through bogofilter -S/-N, but
Im not sure, because that could stop bogofilter to
properly correct the statistics when these headers are
in the original message.
Another option could be to add an option to use with
-S/-N only to let the user tell to bogofilter do not
include the X-Bogosity line in his analysis.
But perhaps the better now is to wait until we have an
integrated tool to manage the learning and correct the
bogofilter mistakes, to gain better feeling about which
is the best solution. After all, this has almost none
(if any) statistical meaning...!