From nix@esperi.demon.co.uk  Fri Jul 19 21:51:59 2002
Return-Path: <nix@esperi.demon.co.uk>
Delivered-To: yyyy@localhost.netnoteinc.com
Received: from localhost (localhost [127.0.0.1])
	by phobos.labs.netnoteinc.com (Postfix) with ESMTP id 6F7B2440C8
	for <jm@localhost>; Fri, 19 Jul 2002 16:51:59 -0400 (EDT)
Received: from dogma.slashnull.org [212.17.35.15]
	by localhost with IMAP (fetchmail-5.9.0)
	for jm@localhost (single-drop); Fri, 19 Jul 2002 21:51:59 +0100 (IST)
Received: from esperi.demon.co.uk (0@esperi.demon.co.uk [194.222.138.8])
    by dogma.slashnull.org (8.11.6/8.11.6) with ESMTP id g6JKqLJ02381 for
    <jm@jmason.org>; Fri, 19 Jul 2002 21:52:21 +0100
Received: from amaterasu.srvr.nix (0@amaterasu.srvr.nix [192.168.1.14]) by
    esperi.demon.co.uk (8.11.6/8.11.6) with ESMTP id g6JKpog31117;
    Fri, 19 Jul 2002 21:51:50 +0100
Received: (from nix@localhost) by amaterasu.srvr.nix (8.11.6/8.11.4) id
    g6JKpnp06336; Fri, 19 Jul 2002 21:51:49 +0100
Sender: nix@esperi.demon.co.uk
To: yyyy@spamassassin.taint.org (Justin Mason)
Cc: Shane Williams <shanew@shanew.net>,
	spamassassin-talk@lists.sourceforge.net
Subject: Re: [SAtalk] Two rule suggestions
References: <20020719110444.E7CD444073@phobos.labs.netnoteinc.com>
X-Emacs: the prosecution rests its case.
From: Nix <nix@esperi.demon.co.uk>
Date: 19 Jul 2002 21:51:49 +0100
In-Reply-To: <20020719110444.E7CD444073@phobos.labs.netnoteinc.com>
Message-Id: <874revl9ui.fsf@amaterasu.srvr.nix>
User-Agent: Gnus/5.0808 (Gnus v5.8.8) XEmacs/21.4 (Informed Management (RC1))
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

On Fri, 19 Jul 2002, Justin Mason mused:
> Shane Williams said:
>> Also, if a single line of yelling scores -0.036, why not just round
>> it off to 0 and not have the test run at all?
> 
> good question ;)

It seems reasonable to have a configurable minimum below which
spamassassin flattens the test to zero (turning it off): by default set
to zero; but if you set it to say 0.01, spamassassin would ignore tests
with scores -0.01<0<0.01 (trading accuracy for time).

The overkill variant of this feature would have scores scored by (score
* relative-time-taken) with relative-time-taken computed over a large
corpus (probably Craig's ;} ). Then you could specify a minimum value of
*that* parameter to flatten at, giving assurances that you're killing
only tests that are expensive for their utility.

Unfortunately I can't think of a good name for that parameter.

I can whip up a patch if anyone thinks this is any use.

-- 
`There's something satisfying about killing JWZ over and over again.'
                                        -- 1i, personal communication