Some tests with the phrases before the dash in the parentheses:

ceo not assistant - brings back jobs with assistant in the title
ceo - works fine
not assistant - brings back 220,000 out of a possible 233,000 which
sounds about right.
ceo not ceo - brings back nothing as expected
ceo not the - brings back jobs with the in the body text. Is it a
default stopword? Just guessing here. 



On Fri, 2005-02-11 at 15:25 -0600, Peter Karman wrote:
> does
> 
> 'job=(ceo) and job=(not assistant)'
> 
> also fail?
> 
> Mark Maunder wrote on 2/11/05 3:22 PM:
> > Gulp. My mistake. Not appears to be completely broken for me.
> > RankScheme(0) just moved the erroneous results further down the pile.
> > Has anyone else experienced a broken NOT when indexing XML and using
> > MetaNames to query?
> > 
> > Thanks.
> > 
> > On Fri, 2005-02-11 at 13:09 -0800, Mark Maunder wrote:
> > 
> >>Hi,
> >>
> >>Firstly tip of the hat to the swish team - the new site rocks, and as
> >>usual so does the little miracle that it supports. 
> >>
> >>I'm indexing using XML with metanames and
> >>IgnoreTotalWordCountWhenRanking no
> >>So when I do a query I'll do something like:
> >>job=(CEO not assistant)
> >>
> >>It looks like NOT is not notting when using RankScheme(1). I'm switching
> >>back to RankScheme(0) for now but I'm going to miss RankScheme(1)
> >>because the sizes of the chunks of text that I index vary wildly and I
> >>find that the larger chunks float to the top with (0). 
> >>
> >>Regards,
> >>
> >>Mark.
> >>
> >>
> >>
>