https://bz.apache.org/SpamAssassin/show_bug.cgi?id=7351

RW <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[email protected]

--- Comment #2 from RW <[email protected]> ---
I think there is a problem here. For a while this month the razor score was
above 4.5 points, and at that level a razor FP combines with BAYES_50 to
produce a misclassification at the default threshold.  The current rules are
over-complicated and make it hard to put a sensible limit on the score.

Engine 4 fell-off a long time ago leaving only engine 8.
RAZOR2_CF_RANGE_E8_51_100 and RAZOR2_CF_RANGE_51_100 are now duplicates, so
RAZOR2_CF_RANGE_E4_51_100 and RAZOR2_CF_RANGE_E8_51_100 are now both
superfluous.

Engine 8 produces mostly high confidence results, in the mass checks
RAZOR2_CHECK only hits 0.3% more spam (for 44% more ham).  Its precise score
wont be all that important - provided it's low. 

What I would suggest is something like:

- Remove RAZOR2_CF_RANGE_E8_51_100 and RAZOR2_CF_RANGE_E4_51_100

- Fix the score for  RAZOR2_CHECK at  0 1.7 0 0.8

- Limit the score for RAZOR2_CF_RANGE_51_100 to 2.7

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to