can we train one that works for many detectors our of the box?
The idea is: try to learn a composable building block (hence LEGO/BRICKS :) ) of next-particle prediction that distills the collective effects of particle interacting with matter that can be iterated just like mechanistic simulators
