Lamarr dependencies for Run 3 productions
As discussed in the June 2022 software week parallel, see slides here, Lamarr will need to be able to create Run 3 style reconstruction or event model objects which can then be read by analysis code. This brings up the question of how best to handle emulating HLT1 for Run 3 Lamarr, since the relevant event model objects live in Allen and not in LHCb.
While it could be envisaged to work around this by making Lamarr emulate HLT1 decisions/efficiencies directly, as opposed to reconstruction or event model objects, this runs into the problem that HLT1 selection efficiencies are necessarily sample dependent (even as a function of kinematics) in a way that single-particle reconstruction efficiencies are not. And the samples, particularly data samples, available to train such decisions are far smaller than the essentially infinite calibration samples available for single-particle reconstruction efficiencies.
We should understand what the concrete options and their pros and cons are in order to take an informed decision on how to proceed.