Using ze-filter headers (filtering on MUA)

The content of this page can help you to explain your users how to exploit information presented on ze-filter headers.

Introduction

When ze-filter checks the content of a message it can add two kind of headers : X-ze-filter-Score and X-ze-filter-Status.

These headers can be used, by the final recipient, to create a filter on the MUA (Thunderbird, …) or a Sieve filter on a Cyrus IMAP server, to redirect the messages to some folders upon some criteria.

These headers look like below :

X-ze-filter-Score: MSGID : 45E477DA.000 on paris-2 : ze-filter score : XXXXX : R=. U=. O=# B=0.999 -> G=5
X-ze-filter-Status: HI

X-ze-filter-Score header gives detailed scores for each filtering class : Pattern Matching, URL filtering, Heuristic filter and Bayesian filter, while X-ze-filter-Status has only two possible values (HI and LO) resuming a global score assigned to the message.

As long as error probability is never zero, the probably best decision, if the score seems to indicate a spam message, is to redirect it to a quarantine folder instead of discarding, rejecting it¹⁾ . Then you can have a look on it for debugging purpose.

Simplified Filtering

Simplified filtering uses the X-ze-filter-Status header, which can have only two values HI and LO.

Usually, filter based on this header will realize following algorithm :

It this header contains HI, then move message to PROBABLY-SPAM folder
If this header contains LO, then move message to SUSPECT folder
Else, move message to Inbox folder

Most of the time, this filter will satisfy most users.

ze-filter sets the content of this header based on the value of two configuration options : XSTATUS_HEADER_HI_REGEX and XSTATUS_HEADER_LO_REGEX. Here is what we have at our servers presently.

XSTATUS_HEADER_HI_REGEX    (U=####|B=0.9|B=0.8|XXXX.*B=0.7)

This means : Header X-ze-filter-Status is set to HI if just added X-ze-filter-Score header matches one of below conditions :

URL filtering score is greater than 3
Bayesian filter score is greater or equal to 0.8
Bayesian filter score is greater or equal to 0.7 AND global score is greater than 3

XSTATUS_HEADER_LO_REGEX    (XXXX|X.*B=0.7|XX.*B=0.6|XXX.*B=0.5)

This means : Header X-ze-filter-Status is set to LO if just added X-ze-filter-Score header matches one of below conditions :

Global score is greater than 3
Bayesian filter score is greater or equal to 0.5 AND global score is greater than 2
Bayesian filter score is greater or equal to 0.6 AND global score is greater than 1
Bayesian filter score is greater or equal to 0.7 AND global score is greater than 0

Advantages of the simplified filter are :

easier to define by most users
thresholds updated on the mail server filter are automatically propagated to user filters

Advanced filtering

In order to use all scores, user should acquire some knowledge about ze-filter filtering and the meaning of each entry in the X-ze-filter-Score header :

X-ze-filter-Score: MSGID : 45E477DA.000 on paris-2 : ze-filter score : XXXXX : R=. U=. O=# B=0.999 -> G=5
                                                                         |      |   |   |   |          |
                                                        Global Score   <-|      |   |   |   |          |
                                              Pattern Matching Score   <--------|   |   |   |          |
                                                 URL Filtering Score   <------------|   |   |          |
                                           Heuristic Filtering Score   <----------------|   |          |
                                            Bayesian Filtering Score   <--------------------|          |
                                                        Global Score   <-------------------------------|

Most of the time, all these individual checks can be combined in a “near binary” classification function to decide if the message is a spam or not. Advanced users can base the initial configuration of their filter on regular expressions (see Simple Filter above) used by ze-filter to set X-ze-filter-Status header value and tweak these expressions to best fit his needs.

¹⁾

As we are operating in before queue mode, we can reject, instead of discarding

Table of Contents

Using ze-filter headers (filtering on MUA)

Introduction

Simplified Filtering

Advanced filtering