Try to pick a representative file when testing
Currently the CI picks a file at random which can cause weird test results due to it being abnormally large/small. A better approach would be to choose a file which is "typical". An example of a test which a bad file choice is:
https://lhcb-analysis-productions.web.cern.ch/2151747/Lb2Lll/data_Collision12_MagDown_tuple/
Edited by Chris Burr