References of interest:

• Les Houches proceedings 2015: https://arxiv.org/abs/1605.04692 (Section III.3: STXS introduction and stage-0 and -1 bin definitions)(identical to what is in YR4)
• WG1 meeting on ggF theoretical uncertainties in kinematic regions: https://indico.cern.ch/event/618048/
• YR4: https://arxiv.org/abs/1610.07922 (I.4.2.a feeds into the uncertainty scheme used for ggF, I.4.2.c gives the resummed ST uncertainties that are the first few parameters in the ggF uncertainty scheme)
• Presentations on non-Gaussian likelihood parameterizations: talk 1 , talk 2 ,

• Slides from Nicolas (STXS)
• Slides from Pedja (differential xs)

Goal of the session: identify topics for detailed discussion

• presentation of the experimental results (which information to make public and how)
• interpretation of STXS results (what is needed from experiment and theory for a simple interpretations such as STXS → production mode signal strength by someone outside of the experiments)
• merging of STXS bins (in case of no experimental sensitivity to a given bin, and/or large correlations. Do we want to propose a “stage-0.5”?)

• additional variables of interest for differential measurements
• how to estimate uncertainties related to model dependence (assuming SM in the calculation of acceptance corrections)

• parametrization of theory uncertainties for VBF and VH (identification of uncorrelated sources in the predictions)
• treatment of gg→Z(→qq)H (currently this is absorbed into gg→H, should it stay like this? should we define dedicated STXS bins?)
• ttH: introduction of dedicated STXS bins?

Consider a simple interpretation starting from the measured STXS and their uncertainties in terms of inclusive signal strength per production process (as a working example – no impact on what is written below).

What is the minimum information needed to be made public? Several options:

First case: everything can be considered Gaussian:

• central values of the measured STXS
• covariance matrices separately for statistical and experimental uncertainties, one covariance matrix for every theoretical uncertainty that one might want to correlate between the measurement and the interpretation, one covariance matrix for all remaining theoretical uncertainties
  • in this case, all NP are profiled and the matrices contain the impact of the uncertainties on the STXS
  • correlation between uncertainties in the measurement and the interpretation is only possible if the corresponding covariance matrices look like (Delta1^2 Delta1Delta2, Delta1Delta2 Delta2^2) (this would have to be checked)
• for uncertainties one might want to correlate between measurement and interpretation, also the size of the uncertainties considered needs to be reported (e.g. the size of the alphas variation considered)
• a somewhat finer splitting of the uncertainties could be useful, e.g. a split into uncertainties related to signal and to background (to be considered case-by-case)

Second case: keep some nuisance parameters unprofiled, but still everything can be considered Gaussian:

• give the covariance matrices with impact of the profiled nuisance parameters on the STXS as above, and extend the vector of central values and the covariance matrices by the unprofiled parameters
• this allows to share nuisance parameters between the measurement and interpretation also if the impact on the STXS does not have the simplest form as above
• rest as above

Non-Gaussian case: uncertainties or one or few STXS cannot be considered Gaussian:

• provide the full likelihood for those STXS, and the information as above for the others