Comments from Andy Buckley:

(Apologies for the incoherence of these: they're essentially a
stream-of-consciousness elaboration on the notes I had as an aide memoire during
the discussion session!)

Generally agree that efforts to coordinate MC tuning to LHC data between LHC
experiments and including theorists & MC authors are welcome. Of course,
experiments, MC authors and tuners will also do their own tunes, but it would be
better if the more generic tunes can be done in common, leaving more time for
more specific studies in more experiment or MC-specific groups.

We have been rather Pythia-specific throughout the workshop: bear in mind the
re-activation of PHOJET, plus new models in Herwig++, SHERPA, ... and others?

Involvement with MC authors is crucial if the data in any significant way
disagree with the models: this is a naturally iterative process, since a tune
must be seriously attempted to discover whether or not the model is capable of a
description, the model must be enhanced/fixed if it cannot, and then a retune
must be attempted... and so on, potentially ad infinitum. We've found that tools
like Rivet and Professor can really help in reducing the turn-around time in
these iterations, as does a close working relationship between tuners and
authors. This has been exactly the case in the work of some Professor authors on
a tune of SHERPA hadronisation.

Professor isn't necessarily a fully-automated tool: full-automation sceptics may
wish to try out the Professor interactive mode, in which a manual tune can be
performed on the parameterisation of generator responses to parameter
changes. This has proven very useful when trying to understand why a good tune
cannot be found automatically.

Rivet 1.2.0 was released during the MB/UE workshop, and a well-defined version
of Professor will be produced for installation as part of the LCG AA/GENSER
suite. We appreciate the suggestion that LCG can help with technical maintenance
of these installations for the community to use!

Competition/complementarity between Professor and PROFIT as tuning tools is
welcomed: in fact the complementarity is genuine in the case of error
estimation, as Professor uses an approach similar to that of the NNPDF
collaboration in contrast to PROFIT's CTEQ-like eigen-decompostion. Developments
in this area will be interesting: in particular there is not yet a way to
statistically generate intuitive error tunes such as Peter Skands' "hard/soft"
Perugia tunes.

Reduction of generator combinatorics is welcome: with all the potential add-ons
for specific decays etc., one could generate a huge number of discrete generator
configurations to be tuned. Fortunately, most are special features such as QED
or tau/B decays and can be fairly easily factorised into mini-tunes. Uncertainty
over the effect of multi-leg matching schemes (e.g. CKKW, MLM) seems to have
been largely resolved by the studies of P. Lenzi, which show that HERWIG and
PYTHIA MB/UE tunes are effectively unchanged by addition of higher-pT matching
schemes.

All experiments have something to add to the tunes: ALICE and LHCb have better
particle ID and lower tracking pT cutoffs than the GPDs, LHCb has a distinctive
pseudorapidity range and an interest in excited mesonic states which have so far
been largely untuned (in any standard tunes of any generator I know of.) In all
cases it will be interesting to test the limits of the process-independence
assumptions in the MC UE models.

To make this tuning possible and the data archiving as robust as possible,
please ensure that your plot data gets archived and submitted to HepData
(http://hepdata.cedar.ac.uk) in a suitable plain text format -- with explicit
bin edges, please! -- this will make export of reference data to tools like
Rivet much easier.

Finally, please try out Rivet and Professor and report your bugs, feature
requests, comments, frustrations etc. to the developers of each: we'll try to
help!