This effort started ~10 years ago, but not much push for it until now
This is a first case study of its feasibility with H->4l events from CMS and ATLAS
We know STXS (in production) is affected by potentially large acceptance effects
If we define Lorentz invariant fiducial regions in decay, we can mitigate them. Lorentz invariance is crucial to reduce DoFs.
The goal is to have the same behaviour (or very close behaviour) between SM and BSM
Detector effects are not completely removable, but we can mitigate them.

Action items:

It would be nice to see the double ratio
It would be nice to compare ZZ* vs gamma*gamma* vs Zgamma*
A: the eigenvectors shown are in any case taken among the most extreme cases
It would be nice to compare ttH and ggH
A: this was checked, the behaviour was the same

The agreement seems worst than in ATLAS case
A: it could be the CMS samples are more extreme, but most likely the fiducial volume was defined fine tuning it on the ATLAS selection and thus CMS pays a price for this. Some work is needed to adapt and compromise the fiducial selection across the 2 experiments.

The 2 experiments need to work a bit more on the definition of the fiducial selection
Using the same samples should help making an apples-to-apples comparison
Material could be shared on the WG2 twiki
If the method works, the idea is to replace the decay defnition in STXS with the result of this study. To be noted: since the fiducial decay volume is not dependent on the production (Lorentz invariance), if we define M STXS production bins and N decay bins, the total number of bins is N+M (and not N*M). The (relative) number of events in each decay bin is the same for each production bin. So stage 0 would be to replace the definition of the decay final state, stage 1 we can introduce some binning (for example mZ2 which is sensitive to CP).