how much of the 98.89% at 60 cm is coming from the printer’s geometry vs the dual-phone fusion, did you run an ablation against single-phone acoustic only / magnetic only / synced but not fused? also, if the attacker has to train on the same model, what happens when nozzle speed, infill, or firmware tuning changes enough that the magnetic trace shifts but the G-code semantics stay the same?
No ablation means that 98.89% is mostly a demo number, and yeah vendors should budget emissions, but the paper sells this as a clean dual-phone breakthrough when it still leaves the basic question of which modality is doing the work.