From gwatts@phys.washington.edu Mon May  3 14:25:05 2004
Date: Fri, 23 Apr 2004 09:51:39 -0700
From: Gordon Watts <gwatts@phys.washington.edu>
To: Gustaaf Brooijmans <gusbroo@fnal.gov>
Cc: Marco Verzocchi <mverzocc@fnal.gov>, d0dfwg@fnal.gov
Subject: RE: Comments from the WZ group on data format issues

Hi,
  It is worse that getting n-1 tuning out of reco. Indeed, they will get
some sort of results out of that, but not only will they not be the most
efficient (perhaps), but the TRF's supplied by the bid group will be
wrong. The results would be good enough for establishing that the
technical bits of your physics analysis work correctly, but not much
more.

  After I sent this email I realized that it just doesn't matter: Once
you have it running in d0correct, you basically have it running in reco.

	Cheers,
		Gordon.

-----Original Message-----
From: Gustaaf Brooijmans [mailto:gusbroo@fnal.gov] 
Sent: Friday, April 23, 2004 9:48 AM
To: Gordon Watts
Cc: Marco Verzocchi; d0dfwg@fnal.gov
Subject: RE: Comments from the WZ group on data format issues

Hi Gordon,

>   I'm wearing bID convener hat for this email...
>
> >chunks.  As we know, it is unavoidable that some corrections and
> algorithm
> >refinements are applied post-reco however, and it's important that we
> >rationalize the way this is done.
>
> Do you consider running at the d0correct level acceptable as opposed
to
> in reco? The reason is our tuning takes place only after we have large
> samples available to us, which means that most if not all the data
will
> have to be reco-ed before we are ready to freeze the algorithms and
the
> parameters.

Yes and no.  I think our reconstruction software is now reasonably
advanced on the asymptotic curve and a base version of b-tagging should
be
included in reco with a corresponding chunk, and then corresponding info
on the thumbnail.  Clearly, with a new reconstruction version one hopes
to
get some improvement and a new tuning is necessary, which then takes
some
time.  But this leaves the door open for people that want to use the
results straight out of reco, possibly at the cost of some percentage in
efficiency and/or purity.  Others will work on and use a more refined
version which is more appropriate for their analysis.

D0correct is a necessity, but we need to be careful about not making it
a
default stage for running algorithms.  I can think of multiple reasons
for
that: speed, reproducibility, ...


Gustaaf