Strong occupancy dependence of bremsstrahlung recovery
As reported by @rmwillia here, there is a strong dependence of brem recovery performance and occupancy (the more tracks, the worse the resolution). This is because the larger the number of tracks (Long+T-track) the higher the chance an Ecal cluster gets wrongly classified as a charged cluster. Right now (just as in Run 1/2) only neutral-classified clusters are used as input for brem recovery.
Given the known occupancy differences (especially downstream of the magnet) in data versus MC, this effect is worse in data.
This can be mitigated by including charged clusters in the matching process, hence relying solely on a track-cluster brem matching criteria from the signal track.
Note: there is an option that doesn't depend on the track-cluster matching already available and that is using BREMTRACKBASEDENERGY, although one has to correct the electron momenta 'by hand' right now. No 'easy' functor functionality exists for this at the moment.
Some initial studies with MC on including charged clusters are given below
Bremsstrahlung 'HasBrem' fractions versus number of Long tracks (black is neutral-only, blue is all CaloHypos as input)
Electron momentum resolutions (standard deviation of \delta p / p in range [-1,1]) versus number of Long tracks (dashed is neutral-only, solid is all CaloHypos as input)
Specific electron momentum resolution distribution (\delta p / p in range [-1,1]) at high/low occupancy (>150 long tracks and <30 tracks respectively): solid is with both neutral+charged-classified clusters, dashed is just neutral
