[Spambayes] Scoring Confusion

Mark Hammond mhammond at skippinet.com.au
Wed Jun 4 10:10:38 EDT 2003

> OK, I admit it, I'm officially confused :-)

Good - you are now qualified to help work on the product <wink>

> I am running the binary installer version of SpamBayes
> (002) on Windows 2000 Pro, with Outlook 2000.
> At Mark's suggestion from last night, I just looked (for the
> first time) at the Spam Clues for a few messages. First, I
> picked a message that it thought wasn't spam. It was scored
> ".32". For me, this should have shown up in the "Possible
> Spam" folder, since I didn't touch the defaults, which are
> set to >90 move to Spam, >15 move to Unsure.

It is important to remember that subsequent training will affect previous
scores.  Eg, if the mail in question arrived a couple of weeks ago, back
then, it may have had a score close to zero.  However, in the meantime your
training operations will have changed this - but each message will not be
rescored until requested.  The "Spam" field for a message is the score when
it was most recently scored - not how it would be scored now.  So it is
*possible* that what you saw here is correct.

> Then I checked a message that was marked as Spam. It had a
> score of "1". It was moved correctly.
> Then I found another one that was untouched, meaning it
> wasn't spam or unsure, but the score inside was "1" (which,
> of course, was correct!).
> So, it seems that the engine is correctly tagging the
> scores, but somehow, the code isn't recognizing those scores
> for the purposes of filtering.
> Any suggestions? (including telling me to go grab the CVS
> code and just install it manually :-)

The next step is to look in the log file (see troubleshooting.html), and try
and locate a few of these messages.  Each message looked at by SpamBayes
will get an entry in the log.  There are 2 possibilities:

* SpamBayes is simply not seeing some messages.  This can happen in an
Exchange Server environment, and possibly in other cases we are yet to nail

* SpamBayes is seeing the messages, and the log will indicate what action
was taken.  Immediately after seeing what you consider an error, get the
spam clues for that message, and ensure it is the same as the "disposition"
(ie, the "Yes/No/Maybe") shown in the log file.


More information about the Spambayes mailing list