Methods to classify events experimentally according to collision geometry are well established and non-controversial when collisions of large ions are studied. However, high luminosity data from $p/d$+A collisions at RHIC and LHC provided some surprising results that either call for new physics or question the applicability of the established methods of event classification for them.
So far there is no consensus in the community what is the proper model and procedure to determine centrality, how to connect observed event activity with collision geometry in very asymmetric, $p/d$+A collisions in the same sense and with the same accuracy as was done in A+A. We argue that high $p_T$ direct photons offer an {\it a posteriori} test of any method suggested to categorize $p/d$+A -- and in general, very asymmetric -- collisions: the method is only viable if the nuclear modification factor for high $p_T$ direct photons is about unity for all centrality classes.