Identify the cause of the ILD vs FCCee strange-jet tagging performance discrepancy

Determine the cause of the observed performance discrepancy in strange-jet tagging between International Large Detector (ILD) full simulation using Particle Transformer with Comprehensive Particle Identification and with truth particle identification and FCCee Delphes using Particle Transformer, given that FCCee results remain better than ILD even when truth particle identification is used; specifically identify the factor(s) beyond particle identification that produce this difference.

Background

The study applies Particle Transformer (ParT) to jet flavor tagging using ILD full simulation and FCCee Delphes, comparing performance across three-category and six-category classifications. For six-category tagging, particle identification (PID) is critical; ILD uses the CPID framework based on TPC dE/dx and calorimeter time-of-arrival, while FCCee uses dN/dx (cluster counting) and ToF-derived track mass. Despite using truth PID in ILD to remove PID imperfections, the FCCee Delphes setup still exhibits better strange-jet tagging performance than ILD full simulation.

The authors explicitly note that PID differences only partially explain the performance gap, implying additional unexplained factors. Identifying these factors is important to understand detector and algorithmic contributions to strange-tagging performance and to guide further improvements and fair cross-detector comparisons.